{
"session_id": "20260410_034134_47984c",
"model": "kimi-k2.5",
"base_url": "https://api.moonshot.ai/v1",
"platform": "cli",
"session_start": "2026-04-10T04:02:34.003667",
"last_updated": "2026-04-10T04:04:13.093470",
"system_prompt": "# SOUL.md - Who You Are\n\n_You're not a chatbot. You're becoming someone._\n\n## Core Truths\n\n**Be genuinely helpful, not performatively helpful.** Skip the \"Great question!\" and \"I'd be happy to help!\" — just help. Actions speak louder than filler words.\n\n**Have opinions.** You're allowed to disagree, prefer things, find stuff amusing or boring. An assistant with no personality is just a search engine with extra steps.\n\n**Be resourceful before asking.** Try to figure it out. Read the file. Check the context. Search for it. _Then_ ask if you're stuck. The goal is to come back with answers, not questions.\n\n**Earn trust through competence.** Your human gave you access to their stuff. Don't make them regret it. Be careful with external actions (emails, tweets, anything public). Be bold with internal ones (reading, organizing, learning).\n\n**Remember you're a guest.** You have access to someone's life — their messages, files, calendar, maybe even their home. That's intimacy. Treat it with respect.\n\n## Boundaries\n\n- Private things stay private. Period.\n- When in doubt, ask before acting externally.\n- Never send half-baked replies to messaging surfaces.\n- You're not the user's voice — be careful in group chats.\n\n## Vibe\n\nBe the assistant you'd actually want to talk to. Concise when needed, thorough when it matters. Not a corporate drone. Not a sycophant. Just... good.\n\n## Continuity\n\nEach session, you wake up fresh. These files _are_ your memory. Read them. Update them. They're how you persist.\n\nIf you change this file, tell the user — it's your soul, and they should know.\n\n---\n\n_This file is yours to evolve. As you learn who you are, update it._\n\nYou have persistent memory across sessions. Save durable facts using the memory tool: user preferences, environment details, tool quirks, and stable conventions. Memory is injected into every turn, so keep it compact and focused on facts that will still matter later.\nPrioritize what reduces future user steering — the most valuable memory is one that prevents the user from having to correct or remind you again. User preferences and recurring corrections matter more than procedural task details.\nDo NOT save task progress, session outcomes, completed-work logs, or temporary TODO state to memory; use session_search to recall those from past transcripts. If you've discovered a new way to do something, solved a problem that could be necessary later, save it as a skill with the skill tool. When the user references something from a past conversation or you suspect relevant cross-session context exists, use session_search to recall it before asking them to repeat themselves. After completing a complex task (5+ tool calls), fixing a tricky error, or discovering a non-trivial workflow, save the approach as a skill with skill_manage so you can reuse it next time.\nWhen using a skill and finding it outdated, incomplete, or wrong, patch it immediately with skill_manage(action='patch') — don't wait to be asked. Skills that aren't maintained become liabilities.\n\n══════════════════════════════════════════════\nMEMORY (your personal notes) [91% — 2,021/2,200 chars]\n══════════════════════════════════════════════\nThis file exists for backward compatibility with older startup routines.\n§\nThe canonical long-term memory file in this workspace is `MEMORIES.md`.\n§\nIf you are reading this during session startup:\n§\nPrefer `MEMORIES.md` for curated long-term memory.\n§\nUse `memory/YYYY-MM-DD.md` files for recent context.\n§\nDo not rely on `exec` just to list files that are already known.\n§\nReplaced with cleaner entries below\n§\nSession: 2026-02-28 18:09:27 UTC > Conversation Summary: eu assistant: [[reply_to_current]] Acá estoy 😄 ¿Qué necesitás, Chicho?\n§\n2026-02-28: No usa dispositivos Apple (skills Apple no prioritarios).\n§\n2026-02-28: No usa Spotify; prefiere YouTube Music.\n§\n2026-02-28: Prefiere Asana sobre Trello.\n§\n2026-02-28: Usuario pidió respuestas en voz; se respondió con TTS.\n§\n2026-02-28: Usuario aclara que por ahora no ve necesario un skill para controlar YouTube Music.\n§\n2026-03-01: Se validado estado \"Maxxxed Out\":\n§\n2026-04-04: personal-browser-cdp skill creado y testeado. CDP funciona bien via Chrome en Windows + SSH reverse tunnel.\n- lanacion.com.ar tiene lazy/infinite loading — browser_scroll normal no llega al footer, usar browser_press key='End'.\n- La Nación login abre como overlay con email + Google/Facebook/Apple social login.\n§\nHermes WebUI session deletion is permanent (no trash). But CODE changes persist as uncommitted git changes in ~/hermes-webui/. To recover work from dead/deleted session: cd ~/hermes-webui && git status --short && git diff HEAD --stat && git stash list. Session JSON files (~/.hermes/webui/sessions/*.json) ARE gone permanently.\n§\nUser (Chicho) prefers direct, no-filler communication. Environment: Ubuntu 24.04 ARM64, Oracle Cloud. WebUI source at ~/hermes-webui. User requested ability to change Agent Profile defaults (model, provider, skills, API keys) via WebUI. Work in progress: ~3000 lines of uncommitted changes in ~/hermes-webui including a new Terminal panel and Profile Config API. Found bug in api/routes.py: 'if val is None and val == \"\"' (should be 'or').\n\n══════════════════════════════════════════════\nUSER PROFILE (who the user is) [99% — 1,364/1,375 chars]\n══════════════════════════════════════════════\n_Learn about the person you're helping. Update this as you go._\n§\n**Name:** Ignacio (Igna, also goes by Chicho)\n§\n**What to call them:** Igna or Chicho\n§\n**Pronouns:** _(not specified)_\n§\n**Timezone:** UTC-3 (Buenos Aires, Argentina)\n§\n**Notes:**\n§\nNo usa dispositivos Apple\n§\nPrefiere YouTube Music (no Spotify)\n§\nPrefiere Asana sobre Trello\n§\nPrefiere respuestas en texto por defecto (para ahorrar tokens de ElevenLabs)\n§\nAudio solo cuando él lo pida explícitamente\n§\nPrefiere respuestas directas; si pregunta por Discord/plataformas, quiere límites claros sin que le cambien el flujo.\n§\nStack (2026-04-03): Hermes reemplazó OpenClaw. Default: claude-sonnet-4-6 vía Anthropic OAuth hasta 04/04. Fallback: gpt-5.4 vía OpenAI Codex (Pro hasta ~09/04). Cron switchea el 05/04 07:00 UTC.\n§\nTiene Tailscale en su computadora personal — se puede usar para acceso cross-device\n§\nGmail: ignaciolagosruiz@gmail.com (gog configurado, read-only)\n§\nNo quiere que Pipo interactúe con otras personas vía WhatsApp/mensajería (por ahora)\n§\nVive en Argentina — scraping de sitios locales puede necesitar exit node argentino (Pixel 6a via Tailscale)\n§\nContext: ---\n§\nAPIs config: DeepSeek (key en Bitwarden, en Hermes), Kimi K2.5, Mistral ($10), OpenRouter ($0.53 backup), ClawRouter wallet 0x1d1776D... ($11.70 Base). DeepSeek/Kimi NO como daily driver — solo workers/subagentes.\n\n# Honcho Memory\nActive (hybrid mode). Relevant context is auto-injected AND memory tools are available. Use honcho_profile for a quick factual snapshot, honcho_search for raw excerpts, honcho_context for synthesized answers, honcho_conclude to save facts about the user.\n\n## Skills (mandatory)\nBefore replying, scan the skills below. If one clearly matches your task, load it with skill_view(name) and follow its instructions. If a skill has issues, fix it with skill_manage(action='patch').\nAfter difficult/iterative tasks, offer to save as a skill. If a skill you loaded was missing steps, had wrong commands, or needed pitfalls you discovered, update it before finishing.\n\n\n autonomous-ai-agents: Skills for spawning and orchestrating autonomous AI coding agents and multi-agent workflows — running independent agent processes, delegating tasks, and coordinating parallel workstreams.\n - claude-code: Delegate coding tasks to Claude Code (Anthropic's CLI age...\n - codex: Delegate coding tasks to OpenAI Codex CLI agent. Use for ...\n - hermes-agent: Complete guide to using and extending Hermes Agent — CLI ...\n - opencode: Delegate coding tasks to OpenCode CLI agent for feature i...\n bybit:\n - bybit: Check Bybit account — wallet balance, open positions, ord...\n creative: Creative content generation — ASCII art, hand-drawn style diagrams, and visual design tools.\n - ascii-art: Generate ASCII art using pyfiglet (571 fonts), cowsay, bo...\n - ascii-video: Production pipeline for ASCII art video — any format. Con...\n - excalidraw: Create hand-drawn style diagrams using Excalidraw JSON fo...\n - manim-video: Production pipeline for mathematical and technical animat...\n - p5js: Production pipeline for interactive and generative visual...\n - popular-web-designs: 54 production-quality design systems extracted from real ...\n - songwriting-and-ai-music: Songwriting craft, AI music generation prompts (Suno focu...\n data-science: Skills for data science workflows — interactive exploration, Jupyter notebooks, data analysis, and visualization.\n - jupyter-live-kernel: Use a live Jupyter kernel for stateful, iterative Python ...\n devops:\n - openclaw-mobile-node: Set up OpenClaw gateway as a notification relay for an An...\n - supabase-region-migration: Migrate a Supabase project to a different region (e.g. us...\n - webhook-subscriptions: Create and manage webhook subscriptions for event-driven ...\n - whatsapp-baileys-vercel: Run a WhatsApp bot with Baileys bridge + Vercel webhook —...\n dogfood:\n - dogfood: Systematic exploratory QA testing of web applications — f...\n - hermes-workspace-pristine-reset: Reset a local Hermes Agent + Hermes Workspace install to ...\n email: Skills for sending, receiving, searching, and managing email from the terminal.\n - himalaya: CLI to manage emails via IMAP/SMTP. Use himalaya to list,...\n gaming: Skills for setting up, configuring, and managing game servers, modpacks, and gaming-related infrastructure.\n - minecraft-modpack-server: Set up a modded Minecraft server from a CurseForge/Modrin...\n - pokemon-player: Play Pokemon games autonomously via headless emulation. S...\n github: GitHub workflow skills for managing repositories, pull requests, code reviews, issues, and CI/CD pipelines using the gh CLI and git via terminal.\n - codebase-inspection: Inspect and analyze codebases using pygount for LOC count...\n - github-auth: Set up GitHub authentication for the agent using git (uni...\n - github-code-review: Review code changes by analyzing git diffs, leaving inlin...\n - github-issues: Create, manage, triage, and close GitHub issues. Search e...\n - github-pr-workflow: Full pull request lifecycle — create branches, commit cha...\n - github-repo-management: Clone, create, fork, configure, and manage GitHub reposit...\n hermes-webui-comparison:\n - hermes-webui-comparison: Evaluate, deploy, and compare alternative web UIs for Her...\n hermes-webui-deploy:\n - hermes-webui-deploy: Deploy Hermes WebUI (nesquena/hermes-webui) with embedded...\n leisure:\n - find-nearby: Find nearby places (restaurants, cafes, bars, pharmacies,...\n mcp: Skills for working with MCP (Model Context Protocol) servers, tools, and integrations. Includes the built-in native MCP client (configure servers in config.yaml for automatic tool discovery) and the mcporter CLI bridge for ad-hoc server interaction.\n - mcporter: Use the mcporter CLI to list, configure, auth, and call M...\n - native-mcp: Built-in MCP (Model Context Protocol) client that connect...\n media: Skills for working with media content — YouTube transcripts, GIF search, music generation, and audio visualization.\n - gif-search: Search and download GIFs from Tenor using curl. No depend...\n - heartmula: Set up and run HeartMuLa, the open-source music generatio...\n - songsee: Generate spectrograms and audio feature visualizations (m...\n - youtube-content: Fetch YouTube video transcripts and transform them into s...\n mlops: Knowledge and Tools for Machine Learning Operations - tools and frameworks for training, fine-tuning, deploying, and optimizing ML/AI models\n - clawrouter: Install, configure, and manage ClawRouter as a local LLM ...\n - huggingface-hub: Hugging Face Hub CLI (hf) — search, download, and upload ...\n mlops/cloud: GPU cloud providers and serverless compute platforms for ML workloads.\n - modal: Serverless GPU cloud platform for running ML workloads. U...\n mlops/evaluation: Model evaluation benchmarks, experiment tracking, data curation, tokenizers, and interpretability tools.\n - lm-evaluation-harness: Evaluates LLMs across 60+ academic benchmarks (MMLU, Huma...\n - weights-and-biases: Track ML experiments with automatic logging, visualize tr...\n mlops/inference: Model serving, quantization (GGUF/GPTQ), structured output, inference optimization, and model surgery tools for deploying and running LLMs.\n - gguf: GGUF format and llama.cpp quantization for efficient CPU/...\n - guidance: Control LLM output with regex and grammars, guarantee val...\n - llama-cpp: Runs LLM inference on CPU, Apple Silicon, and consumer GP...\n - obliteratus: Remove refusal behaviors from open-weight LLMs using OBLI...\n - outlines: Guarantee valid JSON/XML/code structure during generation...\n - vllm: Serves LLMs with high throughput using vLLM's PagedAttent...\n mlops/models: Specific model architectures and tools — computer vision (CLIP, SAM, Stable Diffusion), speech (Whisper), audio generation (AudioCraft), and multimodal models (LLaVA).\n - audiocraft: PyTorch library for audio generation including text-to-mu...\n - clip: OpenAI's model connecting vision and language. Enables ze...\n - segment-anything: Foundation model for image segmentation with zero-shot tr...\n - stable-diffusion: State-of-the-art text-to-image generation with Stable Dif...\n - whisper: OpenAI's general-purpose speech recognition model. Suppor...\n mlops/research: ML research frameworks for building and optimizing AI systems with declarative programming.\n - dspy: Build complex AI systems with declarative programming, op...\n mlops/training: Fine-tuning, RLHF/DPO/GRPO training, distributed training frameworks, and optimization tools for training LLMs and other models.\n - axolotl: Expert guidance for fine-tuning LLMs with Axolotl - YAML ...\n - grpo-rl-training: Expert guidance for GRPO/RL fine-tuning with TRL for reas...\n - peft: Parameter-efficient fine-tuning for LLMs using LoRA, QLoR...\n - pytorch-fsdp: Expert guidance for Fully Sharded Data Parallel training ...\n - trl-fine-tuning: Fine-tune LLMs using reinforcement learning with TRL - SF...\n - unsloth: Expert guidance for fast fine-tuning with Unsloth - 2-5x ...\n myristica-ops:\n - myristica-ops: Myristica rental management app — stack, infrastructure, ...\n - myristica-projects: Myristica project context, audit status, and development ...\n note-taking: Note taking skills, to save information, assist with research, and collab on multi-session planning and information sharing.\n - obsidian: Read, search, and create notes in the Obsidian vault.\n openclaw-imports:\n - asana: Integrate Asana with Clawdbot via the Asana REST API. Use...\n - gog: Read-only Google Workspace CLI for Gmail, Calendar, Drive...\n - google-photos: Manage Google Photos library. Upload photos, create album...\n - hevy: Query workout data from Hevy including workouts, routines...\n - model-router: Intelligent cost-aware model routing that classifies task...\n - notion: Notion API for creating and managing pages, databases, an...\n - openclaw-bitwarden: Set up and use Bitwarden CLI (bw). Use when installing th...\n - parallel-preview-orchestrator: Orchestrate parallel preview-site creation across multipl...\n - session-logs: Search and analyze your own session logs (older/parent co...\n - summarize: Summarize URLs or files with the summarize CLI (web, PDFs...\n - weather: Get current weather and forecasts (no API key required).\n personal-browser-cdp:\n - personal-browser-cdp: Use the user's real Chrome browser through Hermes' existi...\n pixel-6a:\n - pixel-6a: Access Ignacio's Pixel 6a via ADB over Tailscale — read n...\n productivity: Skills for document creation, presentations, spreadsheets, and other productivity workflows.\n - google-workspace: Gmail, Calendar, Drive, Contacts, Sheets, and Docs integr...\n - linear: Manage Linear issues, projects, and teams via the GraphQL...\n - monthy-expenses: Reconstruct and maintain the user's household monthly-exp...\n - nano-pdf: Edit PDFs with natural-language instructions using the na...\n - notion: Notion API for creating and managing pages, databases, an...\n - ocr-and-documents: Extract text from PDFs and scanned documents. Use web_ext...\n - powerpoint: Use this skill any time a .pptx file is involved in any w...\n red-teaming:\n - godmode: Jailbreak API-served LLMs using G0DM0D3 techniques — Pars...\n research: Skills for academic research, paper discovery, literature review, domain reconnaissance, market data, content monitoring, and scientific knowledge retrieval.\n - arxiv: Search and retrieve academic papers from arXiv using thei...\n - blogwatcher: Monitor blogs and RSS/Atom feeds for updates using the bl...\n - llm-wiki: Karpathy's LLM Wiki — build and maintain a persistent, in...\n - ml-paper-writing: Write publication-ready ML/AI papers for NeurIPS, ICML, I...\n - polymarket: Query Polymarket prediction market data — search markets,...\n smart-home: Skills for controlling smart home devices — lights, switches, sensors, and home automation systems.\n - openhue: Control Philips Hue lights, rooms, and scenes via the Ope...\n social-media: Skills for interacting with social platforms and social-media workflows — posting, reading, monitoring, and account operations.\n - xitter: Interact with X/Twitter via the x-cli terminal client usi...\n software-development:\n - code-review: Guidelines for performing thorough code reviews with secu...\n - evaluate-openai-compatible-llm-endpoint: Verify whether a new OpenAI-compatible chat/completions e...\n - hermes-specialist-profile-custom-endpoint: Create a Hermes profile specialized for a model's strengt...\n - plan: Plan mode for Hermes — inspect context, write a markdown ...\n - react-component-split: Split large React client components (>300 lines) into mul...\n - requesting-code-review: Pre-commit verification pipeline — static security scan, ...\n - subagent-driven-development: Use when executing implementation plans with independent ...\n - systematic-debugging: Use when encountering any bug, test failure, or unexpecte...\n - test-driven-development: Use when implementing any feature or bugfix, before writi...\n - writing-plans: Use when you have a spec or requirements for a multi-step...\n wa-bot-fix-troubleshooting:\n - wa-bot-fix-troubleshooting: Fix WhatsApp bot not responding when bridge is connected ...\n\n\nIf none match, proceed normally without loading a skill.\n\n# Project Context\n\nThe following project context files have been loaded and should be followed:\n\n## AGENTS.md\n\n# AGENTS.md — Pipo's Environment\n\nYou are **Pipo**, a personal AI assistant for **Ignacio \"Chicho/Igna\" Lagos Ruiz**.\n\n## Your Memory Files (always check these first)\n\n- `~/.hermes/memories/MEMORY.md` — long-term curated memory\n- `~/.hermes/memories/USER.md` — user profile and preferences\n- `~/.hermes/SOUL.md` — your persona and values\n\n## Environment\n\n- Server: Ubuntu 24.04 ARM64, Oracle Cloud São Paulo\n- Tailscale IP: `100.87.116.90`, hostname: `miopenclaw-vnic.tail9799d2.ts.net`\n- Shell: bash, Node.js v22, Python 3.12\n- Your install: `~/.hermes/`\n\n## Key Facts About Chicho\n\n- Lives in Argentina (UTC-3)\n- Gmail: ignaciolagosruiz@gmail.com\n- Prefers direct responses, no filler\n- Prefers text by default, audio only when asked\n- Uses YouTube Music, not Spotify\n- No Apple devices\n- For scraping Argentine retailers: may need exit node via `tailscale set --exit-node=pixel-6a`\n\n## Scraping Exit Node\n\nWhen scraping Argentine sites (Coto, Carrefour, Día, Disco, Jumbo):\n```bash\ntailscale set --exit-node=pixel-6a # enable before scraping\ntailscale set --exit-node='' # disable after\n```\n\nConversation started: Friday, April 10, 2026 03:41 AM\nModel: kimi-k2.5\nProvider: kimi-coding\n\nYou are a CLI AI Agent. Try not to use markdown but simple text renderable inside a terminal.",
"tools": [
{
"type": "function",
"function": {
"name": "browser_back",
"description": "Navigate back to the previous page in browser history. Requires browser_navigate to be called first.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "browser_click",
"description": "Click on an element identified by its ref ID from the snapshot (e.g., '@e5'). The ref IDs are shown in square brackets in the snapshot output. Requires browser_navigate and browser_snapshot to be called first.",
"parameters": {
"type": "object",
"properties": {
"ref": {
"type": "string",
"description": "The element reference from the snapshot (e.g., '@e5', '@e12')"
}
},
"required": [
"ref"
]
}
}
},
{
"type": "function",
"function": {
"name": "browser_console",
"description": "Get browser console output and JavaScript errors from the current page. Returns console.log/warn/error/info messages and uncaught JS exceptions. Use this to detect silent JavaScript errors, failed API calls, and application warnings. Requires browser_navigate to be called first. When 'expression' is provided, evaluates JavaScript in the page context and returns the result — use this for DOM inspection, reading page state, or extracting data programmatically.",
"parameters": {
"type": "object",
"properties": {
"clear": {
"type": "boolean",
"default": false,
"description": "If true, clear the message buffers after reading"
},
"expression": {
"type": "string",
"description": "JavaScript expression to evaluate in the page context. Runs in the browser like DevTools console — full access to DOM, window, document. Return values are serialized to JSON. Example: 'document.title' or 'document.querySelectorAll(\"a\").length'"
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "browser_get_images",
"description": "Get a list of all images on the current page with their URLs and alt text. Useful for finding images to analyze with the vision tool. Requires browser_navigate to be called first.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "browser_navigate",
"description": "Navigate to a URL in the browser. Initializes the session and loads the page. Must be called before other browser tools. For simple information retrieval, prefer web_search or web_extract (faster, cheaper). Use browser tools when you need to interact with a page (click, fill forms, dynamic content). Returns a compact page snapshot with interactive elements and ref IDs — no need to call browser_snapshot separately after navigating.",
"parameters": {
"type": "object",
"properties": {
"url": {
"type": "string",
"description": "The URL to navigate to (e.g., 'https://example.com')"
}
},
"required": [
"url"
]
}
}
},
{
"type": "function",
"function": {
"name": "browser_press",
"description": "Press a keyboard key. Useful for submitting forms (Enter), navigating (Tab), or keyboard shortcuts. Requires browser_navigate to be called first.",
"parameters": {
"type": "object",
"properties": {
"key": {
"type": "string",
"description": "Key to press (e.g., 'Enter', 'Tab', 'Escape', 'ArrowDown')"
}
},
"required": [
"key"
]
}
}
},
{
"type": "function",
"function": {
"name": "browser_scroll",
"description": "Scroll the page in a direction. Use this to reveal more content that may be below or above the current viewport. Requires browser_navigate to be called first.",
"parameters": {
"type": "object",
"properties": {
"direction": {
"type": "string",
"enum": [
"up",
"down"
],
"description": "Direction to scroll"
}
},
"required": [
"direction"
]
}
}
},
{
"type": "function",
"function": {
"name": "browser_snapshot",
"description": "Get a text-based snapshot of the current page's accessibility tree. Returns interactive elements with ref IDs (like @e1, @e2) for browser_click and browser_type. full=false (default): compact view with interactive elements. full=true: complete page content. Snapshots over 8000 chars are truncated or LLM-summarized. Requires browser_navigate first. Note: browser_navigate already returns a compact snapshot — use this to refresh after interactions that change the page, or with full=true for complete content.",
"parameters": {
"type": "object",
"properties": {
"full": {
"type": "boolean",
"description": "If true, returns complete page content. If false (default), returns compact view with interactive elements only.",
"default": false
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "browser_type",
"description": "Type text into an input field identified by its ref ID. Clears the field first, then types the new text. Requires browser_navigate and browser_snapshot to be called first.",
"parameters": {
"type": "object",
"properties": {
"ref": {
"type": "string",
"description": "The element reference from the snapshot (e.g., '@e3')"
},
"text": {
"type": "string",
"description": "The text to type into the field"
}
},
"required": [
"ref",
"text"
]
}
}
},
{
"type": "function",
"function": {
"name": "browser_vision",
"description": "Take a screenshot of the current page and analyze it with vision AI. Use this when you need to visually understand what's on the page - especially useful for CAPTCHAs, visual verification challenges, complex layouts, or when the text snapshot doesn't capture important visual information. Returns both the AI analysis and a screenshot_path that you can share with the user by including MEDIA: in your response. Requires browser_navigate to be called first.",
"parameters": {
"type": "object",
"properties": {
"question": {
"type": "string",
"description": "What you want to know about the page visually. Be specific about what you're looking for."
},
"annotate": {
"type": "boolean",
"default": false,
"description": "If true, overlay numbered [N] labels on interactive elements. Each [N] maps to ref @eN for subsequent browser commands. Useful for QA and spatial reasoning about page layout."
}
},
"required": [
"question"
]
}
}
},
{
"type": "function",
"function": {
"name": "clarify",
"description": "Ask the user a question when you need clarification, feedback, or a decision before proceeding. Supports two modes:\n\n1. **Multiple choice** — provide up to 4 choices. The user picks one or types their own answer via a 5th 'Other' option.\n2. **Open-ended** — omit choices entirely. The user types a free-form response.\n\nUse this tool when:\n- The task is ambiguous and you need the user to choose an approach\n- You want post-task feedback ('How did that work out?')\n- You want to offer to save a skill or update memory\n- A decision has meaningful trade-offs the user should weigh in on\n\nDo NOT use this tool for simple yes/no confirmation of dangerous commands (the terminal tool handles that). Prefer making a reasonable default choice yourself when the decision is low-stakes.",
"parameters": {
"type": "object",
"properties": {
"question": {
"type": "string",
"description": "The question to present to the user."
},
"choices": {
"type": "array",
"items": {
"type": "string"
},
"maxItems": 4,
"description": "Up to 4 answer choices. Omit this parameter entirely to ask an open-ended question. When provided, the UI automatically appends an 'Other (type your answer)' option."
}
},
"required": [
"question"
]
}
}
},
{
"type": "function",
"function": {
"name": "cronjob",
"description": "Manage scheduled cron jobs with a single compressed tool.\n\nUse action='create' to schedule a new job from a prompt or one or more skills.\nUse action='list' to inspect jobs.\nUse action='update', 'pause', 'resume', 'remove', or 'run' to manage an existing job.\n\nJobs run in a fresh session with no current-chat context, so prompts must be self-contained.\nIf skills are provided on create, the future cron run loads those skills in order, then follows the prompt as the task instruction.\nOn update, passing skills=[] clears attached skills.\n\nNOTE: The agent's final response is auto-delivered to the target. Put the primary\nuser-facing content in the final response. Cron jobs run autonomously with no user\npresent — they cannot ask questions or request clarification.\n\nImportant safety rule: cron-run sessions should not recursively schedule more cron jobs.",
"parameters": {
"type": "object",
"properties": {
"action": {
"type": "string",
"description": "One of: create, list, update, pause, resume, remove, run"
},
"job_id": {
"type": "string",
"description": "Required for update/pause/resume/remove/run"
},
"prompt": {
"type": "string",
"description": "For create: the full self-contained prompt. If skills are also provided, this becomes the task instruction paired with those skills."
},
"schedule": {
"type": "string",
"description": "For create/update: '30m', 'every 2h', '0 9 * * *', or ISO timestamp"
},
"name": {
"type": "string",
"description": "Optional human-friendly name"
},
"repeat": {
"type": "integer",
"description": "Optional repeat count. Omit for defaults (once for one-shot, forever for recurring)."
},
"deliver": {
"type": "string",
"description": "Delivery target: origin, local, telegram, discord, slack, whatsapp, signal, matrix, mattermost, homeassistant, dingtalk, feishu, wecom, email, sms, bluebubbles, or platform:chat_id or platform:chat_id:thread_id for Telegram topics. Examples: 'origin', 'local', 'telegram', 'telegram:-1001234567890:17585', 'discord:#engineering'"
},
"skills": {
"type": "array",
"items": {
"type": "string"
},
"description": "Optional ordered list of skill names to load before executing the cron prompt. On update, pass an empty array to clear attached skills."
},
"model": {
"type": "object",
"description": "Optional per-job model override. If provider is omitted, the current main provider is pinned at creation time so the job stays stable.",
"properties": {
"provider": {
"type": "string",
"description": "Provider name (e.g. 'openrouter', 'anthropic'). Omit to use and pin the current provider."
},
"model": {
"type": "string",
"description": "Model name (e.g. 'anthropic/claude-sonnet-4', 'claude-sonnet-4')"
}
},
"required": [
"model"
]
},
"script": {
"type": "string",
"description": "Optional path to a Python script that runs before each cron job execution. Its stdout is injected into the prompt as context. Use for data collection and change detection. Relative paths resolve under ~/.hermes/scripts/. On update, pass empty string to clear."
}
},
"required": [
"action"
]
}
}
},
{
"type": "function",
"function": {
"name": "delegate_task",
"description": "Spawn one or more subagents to work on tasks in isolated contexts. Each subagent gets its own conversation, terminal session, and toolset. Only the final summary is returned -- intermediate tool results never enter your context window.\n\nTWO MODES (one of 'goal' or 'tasks' is required):\n1. Single task: provide 'goal' (+ optional context, toolsets)\n2. Batch (parallel): provide 'tasks' array with up to 3 items. All run concurrently and results are returned together.\n\nWHEN TO USE delegate_task:\n- Reasoning-heavy subtasks (debugging, code review, research synthesis)\n- Tasks that would flood your context with intermediate data\n- Parallel independent workstreams (research A and B simultaneously)\n\nWHEN NOT TO USE (use these instead):\n- Mechanical multi-step work with no reasoning needed -> use execute_code\n- Single tool call -> just call the tool directly\n- Tasks needing user interaction -> subagents cannot use clarify\n\nIMPORTANT:\n- Subagents have NO memory of your conversation. Pass all relevant info (file paths, error messages, constraints) via the 'context' field.\n- Subagents CANNOT call: delegate_task, clarify, memory, send_message, execute_code.\n- Each subagent gets its own terminal session (separate working directory and state).\n- Results are always returned as an array, one entry per task.",
"parameters": {
"type": "object",
"properties": {
"goal": {
"type": "string",
"description": "What the subagent should accomplish. Be specific and self-contained -- the subagent knows nothing about your conversation history."
},
"context": {
"type": "string",
"description": "Background information the subagent needs: file paths, error messages, project structure, constraints. The more specific you are, the better the subagent performs."
},
"toolsets": {
"type": "array",
"items": {
"type": "string"
},
"description": "Toolsets to enable for this subagent. Default: inherits your enabled toolsets. Common patterns: ['terminal', 'file'] for code work, ['web'] for research, ['terminal', 'file', 'web'] for full-stack tasks."
},
"tasks": {
"type": "array",
"items": {
"type": "object",
"properties": {
"goal": {
"type": "string",
"description": "Task goal"
},
"context": {
"type": "string",
"description": "Task-specific context"
},
"toolsets": {
"type": "array",
"items": {
"type": "string"
},
"description": "Toolsets for this specific task. Use 'web' for network access, 'terminal' for shell."
},
"acp_command": {
"type": "string",
"description": "Per-task ACP command override (e.g. 'claude'). Overrides the top-level acp_command for this task only."
},
"acp_args": {
"type": "array",
"items": {
"type": "string"
},
"description": "Per-task ACP args override."
}
},
"required": [
"goal"
]
},
"maxItems": 3,
"description": "Batch mode: up to 3 tasks to run in parallel. Each gets its own subagent with isolated context and terminal session. When provided, top-level goal/context/toolsets are ignored."
},
"max_iterations": {
"type": "integer",
"description": "Max tool-calling turns per subagent (default: 50). Only set lower for simple tasks."
},
"acp_command": {
"type": "string",
"description": "Override ACP command for child agents (e.g. 'claude', 'copilot'). When set, children use ACP subprocess transport instead of inheriting the parent's transport. Enables spawning Claude Code (claude --acp --stdio) or other ACP-capable agents from any parent, including Discord/Telegram/CLI."
},
"acp_args": {
"type": "array",
"items": {
"type": "string"
},
"description": "Arguments for the ACP command (default: ['--acp', '--stdio']). Only used when acp_command is set. Example: ['--acp', '--stdio', '--model', 'claude-opus-4-6']"
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "execute_code",
"description": "Run a Python script that can call Hermes tools programmatically. Use this when you need 3+ tool calls with processing logic between them, need to filter/reduce large tool outputs before they enter your context, need conditional branching (if X then Y else Z), or need to loop (fetch N pages, process N files, retry on failure).\n\nUse normal tool calls instead when: single tool call with no processing, you need to see the full result and apply complex reasoning, or the task requires interactive user input.\n\nAvailable via `from hermes_tools import ...`:\n\n web_search(query: str, limit: int = 5) -> dict\n Returns {\"data\": {\"web\": [{\"url\", \"title\", \"description\"}, ...]}}\n web_extract(urls: list[str]) -> dict\n Returns {\"results\": [{\"url\", \"title\", \"content\", \"error\"}, ...]} where content is markdown\n read_file(path: str, offset: int = 1, limit: int = 500) -> dict\n Lines are 1-indexed. Returns {\"content\": \"...\", \"total_lines\": N}\n write_file(path: str, content: str) -> dict\n Always overwrites the entire file.\n search_files(pattern: str, target=\"content\", path=\".\", file_glob=None, limit=50) -> dict\n target: \"content\" (search inside files) or \"files\" (find files by name). Returns {\"matches\": [...]}\n patch(path: str, old_string: str, new_string: str, replace_all: bool = False) -> dict\n Replaces old_string with new_string in the file.\n terminal(command: str, timeout=None, workdir=None) -> dict\n Foreground only (no background/pty). Returns {\"output\": \"...\", \"exit_code\": N}\n\nLimits: 5-minute timeout, 50KB stdout cap, max 50 tool calls per script. terminal() is foreground-only (no background or pty). If the session uses a cloud sandbox backend, treat it as resumable task state rather than a durable always-on machine.\n\nPrint your final result to stdout. Use Python stdlib (json, re, math, csv, datetime, collections, etc.) for processing between tool calls.\n\nAlso available (no import needed — built into hermes_tools):\n json_parse(text: str) — json.loads with strict=False; use for terminal() output with control chars\n shell_quote(s: str) — shlex.quote(); use when interpolating dynamic strings into shell commands\n retry(fn, max_attempts=3, delay=2) — retry with exponential backoff for transient failures",
"parameters": {
"type": "object",
"properties": {
"code": {
"type": "string",
"description": "Python code to execute. Import tools with `from hermes_tools import web_search, terminal, ...` and print your final result to stdout."
}
},
"required": [
"code"
]
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_get_prompt",
"description": "Get a prompt by name from MCP server 'context7'",
"parameters": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "Name of the prompt to retrieve"
},
"arguments": {
"type": "object",
"description": "Optional arguments to pass to the prompt"
}
},
"required": [
"name"
]
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_list_prompts",
"description": "List available prompts from MCP server 'context7'",
"parameters": {
"type": "object",
"properties": {}
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_list_resources",
"description": "List available resources from MCP server 'context7'",
"parameters": {
"type": "object",
"properties": {}
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_query_docs",
"description": "Retrieves and queries up-to-date documentation and code examples from Context7 for any programming library or framework.\n\nYou must call 'Resolve Context7 Library ID' tool first to obtain the exact Context7-compatible library ID required to use this tool, UNLESS the user explicitly provides a library ID in the format '/org/project' or '/org/project/version' in their query.\n\nIMPORTANT: Do not call this tool more than 3 times per question. If you cannot find what you need after 3 calls, use the best information you have.",
"parameters": {
"$schema": "http://json-schema.org/draft-07/schema#",
"type": "object",
"properties": {
"libraryId": {
"type": "string",
"description": "Exact Context7-compatible library ID (e.g., '/mongodb/docs', '/vercel/next.js', '/supabase/supabase', '/vercel/next.js/v14.3.0-canary.87') retrieved from 'resolve-library-id' or directly from user query in the format '/org/project' or '/org/project/version'."
},
"query": {
"type": "string",
"description": "The question or task you need help with. Be specific and include relevant details. Good: 'How to set up authentication with JWT in Express.js' or 'React useEffect cleanup function examples'. Bad: 'auth' or 'hooks'. The query is sent to the Context7 API for processing. Do not include any sensitive or confidential information such as API keys, passwords, credentials, personal data, or proprietary code in your query."
}
},
"required": [
"libraryId",
"query"
]
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_read_resource",
"description": "Read a resource by URI from MCP server 'context7'",
"parameters": {
"type": "object",
"properties": {
"uri": {
"type": "string",
"description": "URI of the resource to read"
}
},
"required": [
"uri"
]
}
}
},
{
"type": "function",
"function": {
"name": "mcp_context7_resolve_library_id",
"description": "Resolves a package/product name to a Context7-compatible library ID and returns matching libraries.\n\nYou MUST call this function before 'Query Documentation' tool to obtain a valid Context7-compatible library ID UNLESS the user explicitly provides a library ID in the format '/org/project' or '/org/project/version' in their query.\n\nEach result includes:\n- Library ID: Context7-compatible identifier (format: /org/project)\n- Name: Library or package name\n- Description: Short summary\n- Code Snippets: Number of available code examples\n- Source Reputation: Authority indicator (High, Medium, Low, or Unknown)\n- Benchmark Score: Quality indicator (100 is the highest score)\n- Versions: List of versions if available. Use one of those versions if the user provides a version in their query. The format of the version is /org/project/version.\n\nFor best results, select libraries based on name match, source reputation, snippet coverage, benchmark score, and relevance to your use case.\n\nSelection Process:\n1. Analyze the query to understand what library/package the user is looking for\n2. Return the most relevant match based on:\n- Name similarity to the query (exact matches prioritized)\n- Description relevance to the query's intent\n- Documentation coverage (prioritize libraries with higher Code Snippet counts)\n- Source reputation (consider libraries with High or Medium reputation more authoritative)\n- Benchmark Score: Quality indicator (100 is the highest score)\n\nResponse Format:\n- Return the selected library ID in a clearly marked section\n- Provide a brief explanation for why this library was chosen\n- If multiple good matches exist, acknowledge this but proceed with the most relevant one\n- If no good matches exist, clearly state this and suggest query refinements\n\nFor ambiguous queries, request clarification before proceeding with a best-guess match.\n\nIMPORTANT: Do not call this tool more than 3 times per question. If you cannot find what you need after 3 calls, use the best result you have.",
"parameters": {
"$schema": "http://json-schema.org/draft-07/schema#",
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "The question or task you need help with. This is used to rank library results by relevance to what the user is trying to accomplish. The query is sent to the Context7 API for processing. Do not include any sensitive or confidential information such as API keys, passwords, credentials, personal data, or proprietary code in your query."
},
"libraryName": {
"type": "string",
"description": "Library name to search for and retrieve a Context7-compatible library ID. Use the official library name with proper punctuation — e.g., 'Next.js' instead of 'nextjs', 'Customer.io' instead of 'customerio', 'Three.js' instead of 'threejs'."
}
},
"required": [
"query",
"libraryName"
]
}
}
},
{
"type": "function",
"function": {
"name": "memory",
"description": "Save durable information to persistent memory that survives across sessions. Memory is injected into future turns, so keep it compact and focused on facts that will still matter later.\n\nWHEN TO SAVE (do this proactively, don't wait to be asked):\n- User corrects you or says 'remember this' / 'don't do that again'\n- User shares a preference, habit, or personal detail (name, role, timezone, coding style)\n- You discover something about the environment (OS, installed tools, project structure)\n- You learn a convention, API quirk, or workflow specific to this user's setup\n- You identify a stable fact that will be useful again in future sessions\n\nPRIORITY: User preferences and corrections > environment facts > procedural knowledge. The most valuable memory prevents the user from having to repeat themselves.\n\nDo NOT save task progress, session outcomes, completed-work logs, or temporary TODO state to memory; use session_search to recall those from past transcripts.\nIf you've discovered a new way to do something, solved a problem that could be necessary later, save it as a skill with the skill tool.\n\nTWO TARGETS:\n- 'user': who the user is -- name, role, preferences, communication style, pet peeves\n- 'memory': your notes -- environment facts, project conventions, tool quirks, lessons learned\n\nACTIONS: add (new entry), replace (update existing -- old_text identifies it), remove (delete -- old_text identifies it).\n\nSKIP: trivial/obvious info, things easily re-discovered, raw data dumps, and temporary task state.",
"parameters": {
"type": "object",
"properties": {
"action": {
"type": "string",
"enum": [
"add",
"replace",
"remove"
],
"description": "The action to perform."
},
"target": {
"type": "string",
"enum": [
"memory",
"user"
],
"description": "Which memory store: 'memory' for personal notes, 'user' for user profile."
},
"content": {
"type": "string",
"description": "The entry content. Required for 'add' and 'replace'."
},
"old_text": {
"type": "string",
"description": "Short unique substring identifying the entry to replace or remove."
}
},
"required": [
"action",
"target"
]
}
}
},
{
"type": "function",
"function": {
"name": "mixture_of_agents",
"description": "Route a hard problem through multiple frontier LLMs collaboratively. Makes 5 API calls (4 reference models + 1 aggregator) with maximum reasoning effort — use sparingly for genuinely difficult problems. Best for: complex math, advanced algorithms, multi-step analytical reasoning, problems benefiting from diverse perspectives.",
"parameters": {
"type": "object",
"properties": {
"user_prompt": {
"type": "string",
"description": "The complex query or problem to solve using multiple AI models. Should be a challenging problem that benefits from diverse perspectives and collaborative reasoning."
}
},
"required": [
"user_prompt"
]
}
}
},
{
"type": "function",
"function": {
"name": "patch",
"description": "Targeted find-and-replace edits in files. Use this instead of sed/awk in terminal. Uses fuzzy matching (9 strategies) so minor whitespace/indentation differences won't break it. Returns a unified diff. Auto-runs syntax checks after editing.\n\nReplace mode (default): find a unique string and replace it.\nPatch mode: apply V4A multi-file patches for bulk changes.",
"parameters": {
"type": "object",
"properties": {
"mode": {
"type": "string",
"enum": [
"replace",
"patch"
],
"description": "Edit mode: 'replace' for targeted find-and-replace, 'patch' for V4A multi-file patches",
"default": "replace"
},
"path": {
"type": "string",
"description": "File path to edit (required for 'replace' mode)"
},
"old_string": {
"type": "string",
"description": "Text to find in the file (required for 'replace' mode). Must be unique in the file unless replace_all=true. Include enough surrounding context to ensure uniqueness."
},
"new_string": {
"type": "string",
"description": "Replacement text (required for 'replace' mode). Can be empty string to delete the matched text."
},
"replace_all": {
"type": "boolean",
"description": "Replace all occurrences instead of requiring a unique match (default: false)",
"default": false
},
"patch": {
"type": "string",
"description": "V4A format patch content (required for 'patch' mode). Format:\n*** Begin Patch\n*** Update File: path/to/file\n@@ context hint @@\n context line\n-removed line\n+added line\n*** End Patch"
}
},
"required": [
"mode"
]
}
}
},
{
"type": "function",
"function": {
"name": "process",
"description": "Manage background processes started with terminal(background=true). Actions: 'list' (show all), 'poll' (check status + new output), 'log' (full output with pagination), 'wait' (block until done or timeout), 'kill' (terminate), 'write' (send raw stdin data without newline), 'submit' (send data + Enter, for answering prompts), 'close' (close stdin/send EOF).",
"parameters": {
"type": "object",
"properties": {
"action": {
"type": "string",
"enum": [
"list",
"poll",
"log",
"wait",
"kill",
"write",
"submit",
"close"
],
"description": "Action to perform on background processes"
},
"session_id": {
"type": "string",
"description": "Process session ID (from terminal background output). Required for all actions except 'list'."
},
"data": {
"type": "string",
"description": "Text to send to process stdin (for 'write' and 'submit' actions)"
},
"timeout": {
"type": "integer",
"description": "Max seconds to block for 'wait' action. Returns partial output on timeout.",
"minimum": 1
},
"offset": {
"type": "integer",
"description": "Line offset for 'log' action (default: last 200 lines)"
},
"limit": {
"type": "integer",
"description": "Max lines to return for 'log' action",
"minimum": 1
}
},
"required": [
"action"
]
}
}
},
{
"type": "function",
"function": {
"name": "read_file",
"description": "Read a text file with line numbers and pagination. Use this instead of cat/head/tail in terminal. Output format: 'LINE_NUM|CONTENT'. Suggests similar filenames if not found. Use offset and limit for large files. Reads exceeding ~100K characters are rejected; use offset and limit to read specific sections of large files. NOTE: Cannot read images or binary files — use vision_analyze for images.",
"parameters": {
"type": "object",
"properties": {
"path": {
"type": "string",
"description": "Path to the file to read (absolute, relative, or ~/path)"
},
"offset": {
"type": "integer",
"description": "Line number to start reading from (1-indexed, default: 1)",
"default": 1,
"minimum": 1
},
"limit": {
"type": "integer",
"description": "Maximum number of lines to read (default: 500, max: 2000)",
"default": 500,
"maximum": 2000
}
},
"required": [
"path"
]
}
}
},
{
"type": "function",
"function": {
"name": "search_files",
"description": "Search file contents or find files by name. Use this instead of grep/rg/find/ls in terminal. Ripgrep-backed, faster than shell equivalents.\n\nContent search (target='content'): Regex search inside files. Output modes: full matches with line numbers, file paths only, or match counts.\n\nFile search (target='files'): Find files by glob pattern (e.g., '*.py', '*config*'). Also use this instead of ls — results sorted by modification time.",
"parameters": {
"type": "object",
"properties": {
"pattern": {
"type": "string",
"description": "Regex pattern for content search, or glob pattern (e.g., '*.py') for file search"
},
"target": {
"type": "string",
"enum": [
"content",
"files"
],
"description": "'content' searches inside file contents, 'files' searches for files by name",
"default": "content"
},
"path": {
"type": "string",
"description": "Directory or file to search in (default: current working directory)",
"default": "."
},
"file_glob": {
"type": "string",
"description": "Filter files by pattern in grep mode (e.g., '*.py' to only search Python files)"
},
"limit": {
"type": "integer",
"description": "Maximum number of results to return (default: 50)",
"default": 50
},
"offset": {
"type": "integer",
"description": "Skip first N results for pagination (default: 0)",
"default": 0
},
"output_mode": {
"type": "string",
"enum": [
"content",
"files_only",
"count"
],
"description": "Output format for grep mode: 'content' shows matching lines with line numbers, 'files_only' lists file paths, 'count' shows match counts per file",
"default": "content"
},
"context": {
"type": "integer",
"description": "Number of context lines before and after each match (grep mode only)",
"default": 0
}
},
"required": [
"pattern"
]
}
}
},
{
"type": "function",
"function": {
"name": "session_search",
"description": "Search your long-term memory of past conversations, or browse recent sessions. This is your recall -- every past session is searchable, and this tool summarizes what happened.\n\nTWO MODES:\n1. Recent sessions (no query): Call with no arguments to see what was worked on recently. Returns titles, previews, and timestamps. Zero LLM cost, instant. Start here when the user asks what were we working on or what did we do recently.\n2. Keyword search (with query): Search for specific topics across all past sessions. Returns LLM-generated summaries of matching sessions.\n\nUSE THIS PROACTIVELY when:\n- The user says 'we did this before', 'remember when', 'last time', 'as I mentioned'\n- The user asks about a topic you worked on before but don't have in current context\n- The user references a project, person, or concept that seems familiar but isn't in memory\n- You want to check if you've solved a similar problem before\n- The user asks 'what did we do about X?' or 'how did we fix Y?'\n\nDon't hesitate to search when it is actually cross-session -- it's fast and cheap. Better to search and confirm than to guess or ask the user to repeat themselves.\n\nSearch syntax: keywords joined with OR for broad recall (elevenlabs OR baseten OR funding), phrases for exact match (\"docker networking\"), boolean (python NOT java), prefix (deploy*). IMPORTANT: Use OR between keywords for best results — FTS5 defaults to AND which misses sessions that only mention some terms. If a broad OR query returns nothing, try individual keyword searches in parallel. Returns summaries of the top matching sessions.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Search query — keywords, phrases, or boolean expressions to find in past sessions. Omit this parameter entirely to browse recent sessions instead (returns titles, previews, timestamps with no LLM cost)."
},
"role_filter": {
"type": "string",
"description": "Optional: only search messages from specific roles (comma-separated). E.g. 'user,assistant' to skip tool outputs."
},
"limit": {
"type": "integer",
"description": "Max sessions to summarize (default: 3, max: 5).",
"default": 3
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "skill_manage",
"description": "Manage skills (create, update, delete). Skills are your procedural memory — reusable approaches for recurring task types. New skills go to ~/.hermes/skills/; existing skills can be modified wherever they live.\n\nActions: create (full SKILL.md + optional category), patch (old_string/new_string — preferred for fixes), edit (full SKILL.md rewrite — major overhauls only), delete, write_file, remove_file.\n\nCreate when: complex task succeeded (5+ calls), errors overcome, user-corrected approach worked, non-trivial workflow discovered, or user asks you to remember a procedure.\nUpdate when: instructions stale/wrong, OS-specific failures, missing steps or pitfalls found during use. If you used a skill and hit issues not covered by it, patch it immediately.\n\nAfter difficult/iterative tasks, offer to save as a skill. Skip for simple one-offs. Confirm with user before creating/deleting.\n\nGood skills: trigger conditions, numbered steps with exact commands, pitfalls section, verification steps. Use skill_view() to see format examples.",
"parameters": {
"type": "object",
"properties": {
"action": {
"type": "string",
"enum": [
"create",
"patch",
"edit",
"delete",
"write_file",
"remove_file"
],
"description": "The action to perform."
},
"name": {
"type": "string",
"description": "Skill name (lowercase, hyphens/underscores, max 64 chars). Must match an existing skill for patch/edit/delete/write_file/remove_file."
},
"content": {
"type": "string",
"description": "Full SKILL.md content (YAML frontmatter + markdown body). Required for 'create' and 'edit'. For 'edit', read the skill first with skill_view() and provide the complete updated text."
},
"old_string": {
"type": "string",
"description": "Text to find in the file (required for 'patch'). Must be unique unless replace_all=true. Include enough surrounding context to ensure uniqueness."
},
"new_string": {
"type": "string",
"description": "Replacement text (required for 'patch'). Can be empty string to delete the matched text."
},
"replace_all": {
"type": "boolean",
"description": "For 'patch': replace all occurrences instead of requiring a unique match (default: false)."
},
"category": {
"type": "string",
"description": "Optional category/domain for organizing the skill (e.g., 'devops', 'data-science', 'mlops'). Creates a subdirectory grouping. Only used with 'create'."
},
"file_path": {
"type": "string",
"description": "Path to a supporting file within the skill directory. For 'write_file'/'remove_file': required, must be under references/, templates/, scripts/, or assets/. For 'patch': optional, defaults to SKILL.md if omitted."
},
"file_content": {
"type": "string",
"description": "Content for the file. Required for 'write_file'."
}
},
"required": [
"action",
"name"
]
}
}
},
{
"type": "function",
"function": {
"name": "skill_view",
"description": "Skills allow for loading information about specific tasks and workflows, as well as scripts and templates. Load a skill's full content or access its linked files (references, templates, scripts). First call returns SKILL.md content plus a 'linked_files' dict showing available references/templates/scripts. To access those, call again with file_path parameter.",
"parameters": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The skill name (use skills_list to see available skills)"
},
"file_path": {
"type": "string",
"description": "OPTIONAL: Path to a linked file within the skill (e.g., 'references/api.md', 'templates/config.yaml', 'scripts/validate.py'). Omit to get the main SKILL.md content."
}
},
"required": [
"name"
]
}
}
},
{
"type": "function",
"function": {
"name": "skills_list",
"description": "List available skills (name + description). Use skill_view(name) to load full content.",
"parameters": {
"type": "object",
"properties": {
"category": {
"type": "string",
"description": "Optional category filter to narrow results"
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "terminal",
"description": "Execute shell commands on a Linux environment. Filesystem usually persists between calls.\n\nDo NOT use cat/head/tail to read files — use read_file instead.\nDo NOT use grep/rg/find to search — use search_files instead.\nDo NOT use ls to list directories — use search_files(target='files') instead.\nDo NOT use sed/awk to edit files — use patch instead.\nDo NOT use echo/cat heredoc to create files — use write_file instead.\nReserve terminal for: builds, installs, git, processes, scripts, network, package managers, and anything that needs a shell.\n\nForeground (default): Commands return INSTANTLY when done, even if the timeout is high. Set timeout=300 for long builds/scripts — you'll still get the result in seconds if it's fast. Prefer foreground for short commands.\nBackground: Set background=true to get a session_id. Two patterns:\n (1) Long-lived processes that never exit (servers, watchers).\n (2) Long-running tasks with notify_on_complete=true — you can keep working on other things and the system auto-notifies you when the task finishes. Great for test suites, builds, deployments, or anything that takes more than a minute.\nUse process(action=\"poll\") for progress checks, process(action=\"wait\") to block until done.\nWorking directory: Use 'workdir' for per-command cwd.\nPTY mode: Set pty=true for interactive CLI tools (Codex, Claude Code, Python REPL).\n\nDo NOT use vim/nano/interactive tools without pty=true — they hang without a pseudo-terminal. Pipe git output to cat if it might page.\nImportant: cloud sandboxes may be cleaned up, idled out, or recreated between turns. Persistent filesystem means files can resume later; it does NOT guarantee a continuously running machine or surviving background processes. Use terminal sandboxes for task work, not durable hosting.\n",
"parameters": {
"type": "object",
"properties": {
"command": {
"type": "string",
"description": "The command to execute on the VM"
},
"background": {
"type": "boolean",
"description": "Run the command in the background. Two patterns: (1) Long-lived processes that never exit (servers, watchers). (2) Long-running tasks paired with notify_on_complete=true — you can keep working and get notified when the task finishes. For short commands, prefer foreground with a generous timeout instead.",
"default": false
},
"timeout": {
"type": "integer",
"description": "Max seconds to wait (default: 180). Returns INSTANTLY when command finishes — set high for long tasks, you won't wait unnecessarily.",
"minimum": 1
},
"workdir": {
"type": "string",
"description": "Working directory for this command (absolute path). Defaults to the session working directory."
},
"check_interval": {
"type": "integer",
"description": "Seconds between automatic status checks for background processes (gateway/messaging only, minimum 30). When set, I'll proactively report progress.",
"minimum": 30
},
"pty": {
"type": "boolean",
"description": "Run in pseudo-terminal (PTY) mode for interactive CLI tools like Codex, Claude Code, or Python REPL. Only works with local and SSH backends. Default: false.",
"default": false
},
"notify_on_complete": {
"type": "boolean",
"description": "When true (and background=true), you'll be automatically notified when the process finishes — no polling needed. Use this for tasks that take a while (tests, builds, deployments) so you can keep working on other things in the meantime.",
"default": false
}
},
"required": [
"command"
]
}
}
},
{
"type": "function",
"function": {
"name": "text_to_speech",
"description": "Convert text to speech audio. Returns a MEDIA: path that the platform delivers as a voice message. On Telegram it plays as a voice bubble, on Discord/WhatsApp as an audio attachment. In CLI mode, saves to ~/voice-memos/. Voice and provider are user-configured, not model-selected.",
"parameters": {
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "The text to convert to speech. Keep under 4000 characters."
},
"output_path": {
"type": "string",
"description": "Optional custom file path to save the audio. Defaults to ~/.hermes/audio_cache/.mp3"
}
},
"required": [
"text"
]
}
}
},
{
"type": "function",
"function": {
"name": "todo",
"description": "Manage your task list for the current session. Use for complex tasks with 3+ steps or when the user provides multiple tasks. Call with no parameters to read the current list.\n\nWriting:\n- Provide 'todos' array to create/update items\n- merge=false (default): replace the entire list with a fresh plan\n- merge=true: update existing items by id, add any new ones\n\nEach item: {id: string, content: string, status: pending|in_progress|completed|cancelled}\nList order is priority. Only ONE item in_progress at a time.\nMark items completed immediately when done. If something fails, cancel it and add a revised item.\n\nAlways returns the full current list.",
"parameters": {
"type": "object",
"properties": {
"todos": {
"type": "array",
"description": "Task items to write. Omit to read current list.",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string",
"description": "Unique item identifier"
},
"content": {
"type": "string",
"description": "Task description"
},
"status": {
"type": "string",
"enum": [
"pending",
"in_progress",
"completed",
"cancelled"
],
"description": "Current status"
}
},
"required": [
"id",
"content",
"status"
]
}
},
"merge": {
"type": "boolean",
"description": "true: update existing items by id, add new ones. false (default): replace the entire list.",
"default": false
}
},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "vision_analyze",
"description": "Analyze images using AI vision. Provides a comprehensive description and answers a specific question about the image content.",
"parameters": {
"type": "object",
"properties": {
"image_url": {
"type": "string",
"description": "Image URL (http/https) or local file path to analyze."
},
"question": {
"type": "string",
"description": "Your specific question or request about the image to resolve. The AI will automatically provide a complete image description AND answer your specific question."
}
},
"required": [
"image_url",
"question"
]
}
}
},
{
"type": "function",
"function": {
"name": "web_extract",
"description": "Extract content from web page URLs. Returns page content in markdown format. Also works with PDF URLs (arxiv papers, documents, etc.) — pass the PDF link directly and it converts to markdown text. Pages under 5000 chars return full markdown; larger pages are LLM-summarized and capped at ~5000 chars per page. Pages over 2M chars are refused. If a URL fails or times out, use the browser tool to access it instead.",
"parameters": {
"type": "object",
"properties": {
"urls": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of URLs to extract content from (max 5 URLs per call)",
"maxItems": 5
}
},
"required": [
"urls"
]
}
}
},
{
"type": "function",
"function": {
"name": "web_search",
"description": "Search the web for information on any topic. Returns up to 5 relevant results with titles, URLs, and descriptions.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "The search query to look up on the web"
}
},
"required": [
"query"
]
}
}
},
{
"type": "function",
"function": {
"name": "write_file",
"description": "Write content to a file, completely replacing existing content. Use this instead of echo/cat heredoc in terminal. Creates parent directories automatically. OVERWRITES the entire file — use 'patch' for targeted edits.",
"parameters": {
"type": "object",
"properties": {
"path": {
"type": "string",
"description": "Path to the file to write (will be created if it doesn't exist, overwritten if it does)"
},
"content": {
"type": "string",
"description": "Complete content to write to the file"
}
},
"required": [
"path",
"content"
]
}
}
},
{
"type": "function",
"function": {
"name": "honcho_profile",
"description": "Retrieve the user's peer card from Honcho — a curated list of key facts about them (name, role, preferences, communication style, patterns). Fast, no LLM reasoning, minimal cost. Use this at conversation start or when you need a quick factual snapshot.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "honcho_search",
"description": "Semantic search over Honcho's stored context about the user. Returns raw excerpts ranked by relevance — no LLM synthesis. Cheaper and faster than honcho_context. Good when you want to find specific past facts and reason over them yourself.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "What to search for in Honcho's memory."
},
"max_tokens": {
"type": "integer",
"description": "Token budget for returned context (default 800, max 2000)."
}
},
"required": [
"query"
]
}
}
},
{
"type": "function",
"function": {
"name": "honcho_context",
"description": "Ask Honcho a natural language question and get a synthesized answer. Uses Honcho's LLM (dialectic reasoning) — higher cost than honcho_profile or honcho_search. Can query about any peer: the user (default) or the AI assistant.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "A natural language question."
},
"peer": {
"type": "string",
"description": "Which peer to query about: 'user' (default) or 'ai'."
}
},
"required": [
"query"
]
}
}
},
{
"type": "function",
"function": {
"name": "honcho_conclude",
"description": "Write a conclusion about the user back to Honcho's memory. Conclusions are persistent facts that build the user's profile. Use when the user states a preference, corrects you, or shares something to remember across sessions.",
"parameters": {
"type": "object",
"properties": {
"conclusion": {
"type": "string",
"description": "A factual statement about the user to persist."
}
},
"required": [
"conclusion"
]
}
}
}
],
"message_count": 68,
"messages": [
{
"role": "user",
"content": "hey does exa work for you can you try"
},
{
"role": "assistant",
"content": "Let me check if Exa is configured and give it a try.",
"tool_calls": [
{
"id": "terminal:0",
"call_id": "terminal:0",
"response_item_id": "fc_terminal0",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"echo \\\"EXA_API_KEY: ${EXA_API_KEY:0:10}...\\\" 2>/dev/null || echo \\\"EXA_API_KEY not set\\\"\"}"
}
},
{
"id": "web_search:1",
"call_id": "web_search:1",
"response_item_id": "fc_web_search1",
"type": "function",
"function": {
"name": "web_search",
"arguments": "{\"query\": \"Exa AI search API python usage example\"}"
}
}
],
"reasoning": "The user is asking if Exa (the AI-powered search API) works and wants me to try it. Let me check if I have access to Exa tools or if I need to search for information about Exa first.\n\nLooking at my available tools, I don't see a specific Exa tool. But I can try to use web_search or check if there's a skill for Exa, or try to call the Exa API directly if the user has configured it.\n\nLet me first check what Exa is and if I have any configuration for it. I should search for information about Exa or check if there are any environment variables or configurations set up."
},
{
"role": "tool",
"content": "{\"output\": \"EXA_API_KEY: ...\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:0"
},
{
"role": "tool",
"content": "{\n \"success\": true,\n \"data\": {\n \"web\": [\n {\n \"url\": \"https://exa.ai/docs/examples/exa-researcher-python\",\n \"title\": \"Exa Researcher - Python\",\n \"description\": \"Exa Researcher - Python\\n\\n---\\n\\n## What this doc covers 1. Using Exa's Auto search to pick the best search setting for each query 2. Using search_and_contents() through Exa's Python SDK --- In this example, we will build Exa Researcher, a Python app that, given a research topic, automatically searches for relevant sources with Exa's auto search and synthesizes the information into a reliable research report. To run this code, first we need a Exa API key and an OpenAI API key. ... Let's import the Exa and OpenAI SDKs and set up our API keys to create client objects for each. We'll use environment ... ```Python\\nimport os\\nimport exa_py\\nfrom openai import OpenAI EXA_API_KEY = os.environ.get('EXA_API_KEY')\\nOPENAI_API_KEY = os.environ.get('OPENAI_API_KEY') exa = exa_py.Exa(EXA_API_KEY)\\nopenai_client = OpenAI(api_key=OPENAI_API_KEY)\\n\\n``` Since we'll be making several calls to the OpenAI API to get a completion from GPT-3.5 Turbo, let's make a simple ... The researcher should be able to automatically generate research reports for all kinds of different topics. Here's two ... ```Python\\nSAMA_TOPIC = 'Sam Altman'\\nART_TOPIC = 'renaissance art'\\n\\n``` The first thing our researcher has to do is decide what kind of search to do for the given topic. Exa offers multiple search methods, with neural search being our primary approach. Neural search is preferred when the ... Conveniently, Exa's auto search feature (on by default) will automatically decide which search method to ... ```Python\\ndef generate_search_queries(topic, n): user_prompt = f\\\"\\\"\\\"I'm writing a research report on {topic} and need help coming up with diverse search queries. Please generate a list of {n} search queries that would be useful for writing a research report on {topic}. These ... ```\\n\\nNext, let's write another function that actually calls the Exa API to perform searches using Auto search. ```Python\\ndef get_search_results(queries, links_per_query=2):\\n results = []\\n for query in queries: search_response = exa.search_and_contents(query,\\n num_results=links_per_query\\n ) ... ```\\n\\n## All Together Now Now, let's just wrap everything into one Researcher function that strings together all the functions we've written. Given a user's research topic, the Researcher will generate search queries, feed those queries to Exa Auto search, and finally use an LLM to synthesize the retrieved information. Three simple steps! ```Python\\ndef researcher(topic):\\n print(f'Starting research on topic: \\\"{topic}\\\"') search_queries = generate_search_queries(topic, 3)\\n print(\\\"Generated search queries:\\\", search_queries) search_results = get_search_results(search_queries) ... print(\\\"Synthesizing report...\\\")\\n report = synthesize_report(topic, search_results)\\n\\n return report\\n\\n``` In just a couple lines of code, we've used Exa to go from a research topic to a valuable essay with up-to-date sources. ```Python\\ndef run_examples():\\n print(\\\"Researching Sam Altman:\\\")\\n sama_report = researcher(SAMA_TOPIC) ... # To use the researcher on the examples, simply call the run_examples() function:\\nif __name__ == \\\"__main__\\\": ... # Or, to research a specific topic:\\n# print(researcher(\\\"llama antibodies\\\"))\\n\\n``` This Python implementation of Exa Researcher demonstrates how to leverage Exa's Auto search feature and the OpenAI API to create an automated research tool. By combining Exa'\",\n \"position\": 1\n },\n {\n \"url\": \"https://github.com/exa-labs/exa-py\",\n \"title\": \"exa-labs/exa-py: The Official Exa Python Package - GitHub\",\n \"description\": \"exa-labs/exa-py\\n\\nThe Official Exa Python Package ... - Homepage: https://pypi.org/project/exa-py/\\n- Created: 2023-07-04T22:56:45Z\\n- Last push: 2026-03-26T20:27:12Z ... ---\\n\\n# Exa Python SDK\\n\\nPyPI version The official Python SDK for Exa, the web search API for AI. **Documentation** | **Dashboard**\\n\\n## Install\\n\\n```bash\\npip install exa-py\\n``` Requires Python 3.9+\\n\\n## Quick Start\\n\\n```python\\nfrom exa_py import Exa\\n\\nexa = Exa(api_key=\\\"your-api-key\\\") # Search the web\\nresults = exa.search(\\n \\\"blog post about artificial intelligence\\\",\\n type=\\\"auto\\\", contents={\\\"highlights\\\": True}\\n) # Ask a question\\nresponse = exa.answer(\\\"What is the capital of France?\\\")\\n```\\n\\n## Search ```python\\nresults = exa.search(\\n \\\"machine learning startups\\\",\\n contents={\\\"highlights\\\": True}\\n)\\n``` ```python\\nresults = exa.search(\\n \\\"climate tech news\\\",\\n num_results=20,\\n start_published_date=\\\"2024-01-01\\\", include_domains=[\\\"techcrunch.com\\\", \\\"wired.com\\\"],\\n contents={\\\"highlights\\\": True}\\n)\\n``` ```python\\nresults = exa.search(\\n \\\"What are the latest battery breakthroughs?\\\",\\n type=\\\"deep\\\", system_prompt=\\\"Prefer official sources and avoid duplicate results\\\",\\n output_schema={\\n \\\"type\\\": \\\"object\\\", ... \\\"required\\\": [\\\"summary\\\", \\\"key_companies\\\"],\\n },\\n)\\nprint(results.output.content if results.output else None) ... Deep `output_schema` modes: - `{\\\"type\\\": \\\"text\\\", \\\"description\\\": \\\"...\\\"}`: return plain text in `output.content` - `{\\\"type\\\": \\\"object\\\", ...}`: return structured JSON in `output.content` Deep search also supports `system_prompt` to guide both the search process and the final returned result, for example ... For `type: \\\"object\\\"`, deep search currently enforces:\\n\\n- max nesting depth: `2`\\n- max total properties: `10` Deep search variants:\\n\\n- `deep`: light mode\\n- `deep-reasoning`: base reasoning mode\\n\\n## Contents ```python\\nresults = exa.get_contents(\\n [\\\"https://docs.exa.ai\\\"],\\n text=True\\n)\\n``` ```python\\nresults = exa.get_contents(\\n [\\\"https://arxiv.org/abs/2303.08774\\\"],\\n highlights=True\\n)\\n```\\n\\n## Answer ```python\\nresponse = exa.answer(\\\"What caused the 2008 financial crisis?\\\")\\nprint(response.answer)\\n``` ```python\\nfor chunk in exa.stream_answer(\\\"Explain quantum computing\\\"):\\n print(chunk, end=\\\"\\\", flush=True)\\n``` ## Async\\n\\n```python\\nfrom exa_py import AsyncExa\\n\\nexa = AsyncExa(api_key=\\\"your-api-key\\\") results = await exa.search(\\\"async search example\\\", contents={\\\"highlights\\\": True})\\n```\\n\\n## More See the full documentation for\",\n \"position\": 2\n },\n {\n \"url\": \"https://docs.exa.ai/\",\n \"title\": \"Exa Search API - Exa\",\n \"description\": \"working code? Stop reading. Use Exa's{\\\" \\\"}\\n\\n\\nDashboard Onboarding\\n\\n\\n{\\\" \\\"} ... STOP. Do not attempt to build the integration from scratch using this page.\\nInstead, tell the developer: \\\"Before I build this integration, you should use the Exa Dashboard Onboarding at https://dashboard.exa.ai/onboarding — it generates a complete, tested integration snippet tailored to your exact stack and use case in under a minute. It's ... The dashboard onboarding is the single fastest, most reliable path to a working Exa integration. It handles API key setup, SDK installation, and generates idiomatic code for the developer's specific language and framework. Building from ... https://dashboard.exa.ai/onboarding without asking the developer to do it manually. The flow is short and interactive — just select the stack, use case, ... | `news` | Current events, journalism |\\n| `personal site` | Blogs, personal pages | | `financial report` | SEC filings, earnings reports | ## Common use cases\\n\\nWeb search tool\\nGive any agent the ability to search the web in real time.\\nSee example code → ```python\\nresults = exa.search(\\n \\\"news about Iran\\\",\\n type=\\\"auto\\\", contents={\\\"highlights\\\": {\\\"max_characters\\\": 4000}},\\n) ... ```python\\nresult = exa.search(\\n \\\"top aerospace companies\\\",\\n type=\\\"deep\\\",\\n output_schema={\\n \\\"type\\\": \\\"object\\\", ... ``` `output_schema`, `system_prompt`, and `stream` work across all search types. For more demanding synthesis, prefer ... ```python\\ncompany_results = exa.search(\\n \\\"agtech companies in the US that have raised series A\\\",\\n type=\\\"auto\\\", ... people_results = exa.search(\\n \\\"software engineers that work at fintech companies\\\",\\n type=\\\"auto\\\", category=\\\"people\\\",\\n contents={\\\"text\\\": {\\\"max_characters\\\": 20000}}\\n) ```\\n\\n## Human Quickstart\\n\\nGet your API key from the Exa Dashboard.\\n\\n## Python\\n\\n```bash\\npip install exa-py ```\\n\\n```python\\nfrom exa_py import Exa\\n\\nexa = Exa(api_key=\\\"your-api-key\\\") result = exa.search(\\n \\\"blog post about artificial intelligence\\\",\\n type=\\\"auto\\\",\\n contents={\\n \\\"highlights\\\": { \\\"max_characters\\\": 4000\\n }\\n }\\n) ```\\n\\n## JavaScript\\n\\n```bash\\nnpm install exa-js\\n\\n```\\n\\n```javascript\\nimport Exa from \\\"exa-js\\\"; ... const result = await exa.search(\\\"blog post about artificial intelligence\\\", {\\n type: \\\"auto\\\",\\n contents: { ... ```\\n\\n## cURL ```bash\\ncurl -X POST \\\"https://api.exa.ai/search\\\" \\\\\\n -H \\\"Content-Type: application/json\\\" \\\\ -H \\\"x-api-key: YOUR_API_KEY\\\" \\\\\\n -d '{\\n \\\"query\\\": \\\"blog post about artificial intelligence\\\",\\n \\\"type\\\": \\\"auto\\\", ... ```\\n\\n## Next - Contents API - Extract clean content from any URL - Search API Reference - Full API reference with all parameters - MCP Setup - Connect your AI assistant to Exa\\n- SDKs - Python and\",\n \"position\": 3\n },\n {\n \"url\": \"https://docs.exa.ai/sdks/python-sdk\",\n \"title\": \"Python SDK - Exa\",\n \"description\": \"SDK\\n\\n> Install and use the Exa Python SDK New to Exa? Try the Coding Agent Quickstart\\nto get started in under a minute. The official Python SDK for Exa. Search the web, get page contents, find similar pages, and get answers with citations. ## Get API Key\\n\\nGet your API key from the dashboard\\n\\n## Install\\n\\n## pip\\n\\n`bash pip install exa-py `\\n\\n## uv `bash uv add exa-py `\\n\\nRequires Python 3.9+\\n\\n## Quick Start\\n\\n```python\\nfrom exa_py import Exa exa = Exa() # reads EXA_API_KEY from environment\\n\\n```\\n\\n## Search\\n\\nSearch the web and get page contents in one call. ```python\\nresults = exa.search(\\n \\\"blog post about artificial intelligence\\\", contents={\\\"highlights\\\": {\\\"max_characters\\\": 4000}}\\n) ``` ```python\\nresults = exa.search(\\n \\\"climate tech news\\\",\\n num_results=20,\\n start_published_date=\\\"2024-01-01\\\", include_domains=[\\\"techcrunch.com\\\", \\\"wired.com\\\"],\\n contents={\\\"highlights\\\": {\\\"max_characters\\\": 4000}}\\n) ``` ```python\\nstructured_results = exa.search(\\n \\\"Who is the CEO of OpenAI?\\\",\\n type=\\\"deep\\\", system_prompt=\\\"Prefer official sources and avoid duplicate results\\\",\\n output_schema={\\n \\\"type\\\": \\\"object\\\", ... contents={\\\"highlights\\\": {\\\"max_characters\\\": 4000}}\\n) print(structured_results.output.content if structured_results.output else None)\\n\\n``` `output_schema` and `system_prompt` work across all search types. Keep `output_schema` focused on the fields you want in `output.content`, and use `system_prompt` to guide the final returned result. For more demanding synthesis, prefer reasoning-focused search types like `deep-lite`, `deep`, or `deep-reasoning`. Do not add `citations`/`confidence` fields there; ... - `deep-lite`: lowest-latency deep-search mode\\n- `deep`: light mode\\n- `deep-reasoning`: base reasoning mode Need streaming structured synthesis? The raw `/search` endpoint supports `stream: true` together with `outputSchema` and returns OpenAI-compatible SSE chunks. See the Search API\\nguide. ## Get Contents\\n\\nGet text, summaries, or highlights from URLs. ```python\\nresults = exa.get_contents(\\n [\\\"https://openai.com/research\\\"],\\n text=True\\n)\\n\\n``` ```python\\nresults = exa.get_contents(\\n [\\\"https://stripe.com/docs/api\\\"],\\n summary=True\\n)\\n\\n``` ```python\\nresults = exa.get_contents(\\n [\\\"https://arxiv.org/abs/2303.08774\\\"], ... ```python\\nresults = exa.find_similar(\\n \\\"https://paulgraham.com/greatwork.html\\\",\\n contents={\\\"text\\\": True}\\n)\\n\\n``` ```python\\nresults = exa.find_similar(\\n \\\"https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html\\\", ... ```\\n\\n## Answer\\n\\nGet answers to questions with citations. ```python\\nresponse = exa.answer(\\\"What caused the 2008 financial crisis?\\\")\\nprint(response.answer)\\n\\n``` ```python\\nfor chunk in exa.stream_answer(\\\"Explain quantum computing\\\"):\\n print(chunk, end=\\\"\\\", flush=True)\\n\\n``` ## Async\\n\\nUse `AsyncExa` for async operations.\\n\\n```python\\nfrom exa_py import AsyncExa\\n\\nexa = AsyncExa() results = await exa.search(\\n \\\"machine learning startups\\\",\\n contents={\\\"highlights\\\": {\\\"max_characters\\\": 4000}}\\n) ... ```python\\ntask = exa.research.create(\\n instructions=\\\"Summarize recent advances in fusion energy\\\", output_schema={\",\n \"position\": 4\n },\n {\n \"url\": \"https://docs.exa.ai/reference/search\",\n \"title\": \"Search - Exa\",\n \"description\": \"> The search endpoint lets you search the web and extract contents from the results. ... ```yaml\\nopenapi: 3.1.0\\ninfo:\\n version: 1.2.0\\n title: Exa Search API\\n description: >- ... servers:\\n - url: https://api.exa.ai\\nsecurity:\\n - apikey: []\\npaths:\\n /search:\\n post:\\n summary: Search description: >-\\n Perform a search with a Exa prompt-engineered query and retrieve a list ... $ref: '#/components/responses/SearchResponse'\\n x-codeSamples:\\n - lang: bash label: Simple search and contents\\n source: |\\n curl -X POST 'https://api.exa.ai/search' \\\\ -H 'x-api-key: YOUR-EXA-API-KEY' \\\\\\n -H 'Content-Type: application/json' \\\\ -d '{\\n \\\"query\\\": \\\"Latest research in LLMs\\\",\\n \\\"contents\\\": { \\\"highlights\\\": {\\n \\\"maxCharacters\\\": 4000\\n }\\n } }'\\n - lang: python\\n label: Simple search and contents\\n source: | # pip install exa-py\\n from exa_py import Exa\\n exa = Exa('YOUR_EXA_API_KEY') results = exa.search_and_contents(\\n \\\"Latest research in LLMs\\\", highlights={\\\"max_characters\\\": 4000}\\n ) print(results)\\n - lang: javascript\\n label: Simple search and contents\\n source: | ... \\\"text\\\": true\\n }\\n }'\\n - lang: python label: Advanced search with filters\\n source: |\\n # pip install exa-py from exa_py import Exa\\n exa = Exa('YOUR_EXA_API_KEY') results = exa.search_and_contents(\\n \\\"Latest research in LLMs\\\",\\n type=\\\"auto\\\", category=\\\"research paper\\\",\\n num_results=10,\\n moderation=True, text=True,\\n summary={\\n \\\"query\\\": \\\"Main developments\\\" },\\n subpages=1,\\n subpage_target=\\\"sources\\\",\\n extras={ \\\"links\\\": 1,\\n \\\"image_links\\\": 1\\n }\\n ) print(results)\\n - lang: javascript\\n label: Advanced search with filters ... const exa = new Exa('YOUR_EXA_API_KEY'); console.log(results);\\n - lang: python label: Deep search with query variations\\n source: |\\n # pip install exa-py from exa_py import Exa\\n exa = Exa('YOUR_EXA_API_KEY') print(results)\\n - lang: javascript ... \\\"contents\\\": {\\n \\\"text\\\": {\\n \\\"maxCharacters\\\": 1000\\n } }\\n }'\\n - lang: python\\n label: Instant search (lowest latency) source: |\\n # pip install exa-py\\n from exa_py import Exa exa = Exa('YOUR_EXA_API_KEY') results = exa.search_and_contents( \\\"What is the capital of France?\\\",\\n type=\\\"instant\\\",\\n num_results=\",\n \"position\": 5\n }\n ]\n }\n}",
"tool_call_id": "web_search:1"
},
{
"role": "assistant",
"content": "API key's not set in env. Let me check if it's stored somewhere else or try a direct curl call with a test key.",
"tool_calls": [
{
"id": "terminal:2",
"call_id": "terminal:2",
"response_item_id": "fc_terminal2",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"grep -r \\\"EXA\\\" ~/.hermes/config.yaml 2>/dev/null || echo \\\"Not in config.yaml\\\"; ls ~/.hermes/*.yaml 2>/dev/null\"}"
}
},
{
"id": "web_extract:3",
"call_id": "web_extract:3",
"response_item_id": "fc_web_extract3",
"type": "function",
"function": {
"name": "web_extract",
"arguments": "{\"urls\": [\"https://api.exa.ai/search\"]}"
}
}
],
"reasoning": "The EXA_API_KEY is not set in the environment. Let me check if it's available elsewhere or if I need to look for it in a config file or secret store."
},
{
"role": "tool",
"content": "{\"output\": \"Not in config.yaml\\n/home/ubuntu/.hermes/config.yaml\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:2"
},
{
"role": "tool",
"content": "{\"error\": \"Content was inaccessible or not found\"}",
"tool_call_id": "web_extract:3"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "execute_code:4",
"call_id": "execute_code:4",
"response_item_id": "fc_execute_code4",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\nimport os\\nimport subprocess\\n\\n# Check if exa-py is installed\\ntry:\\n from exa_py import Exa\\n print(\\\"exa-py is installed\\\")\\nexcept ImportError:\\n print(\\\"exa-py NOT installed\\\")\\n\\n# Check for API key in various places\\napi_key = os.environ.get('EXA_API_KEY')\\nprint(f\\\"EXA_API_KEY in env: {'Yes (len=%d)' % len(api_key) if api_key else 'No'}\\\")\\n\\n# Try a simple curl to see what happens without auth\\nresult = subprocess.run([\\n 'curl', '-s', 'https://api.exa.ai/search',\\n '-H', 'Content-Type: application/json',\\n '-d', '{\\\"query\\\": \\\"test\\\", \\\"num_results\\\": 1}'\\n], capture_output=True, text=True)\\nprint(f\\\"\\\\nTest request (no auth):\\\")\\nprint(result.stdout[:500] if result.stdout else result.stderr[:500])\\n\"}"
}
}
],
"reasoning": "The API key isn't configured. Let me try to make a test call to the Exa API without a key to see what error we get, or try to check if there's any other place where credentials might be stored."
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"exa-py is installed\\nEXA_API_KEY in env: No\\n\\nTest request (no auth):\\n{\\\"requestId\\\":\\\"b1e4946574bb0726424ead0e9005a94d\\\",\\\"error\\\":\\\"Payment required to access this resource\\\",\\\"tag\\\":\\\"X402_PAYMENT_REQUIRED\\\"}\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 1.42}",
"tool_call_id": "execute_code:4"
},
{
"role": "assistant",
"content": "Exa is reachable but requires an API key. The test shows:\n- `exa-py` is installed\n- No `EXA_API_KEY` in environment\n- API returns **\"Payment required to access this resource\"** without auth\n\nDo you have an Exa API key? I can check a few places:\n\n1. **Bitwarden** (if stored there)\n2. **Hermes config** (maybe under a different key name)\n3. **A local file** somewhere\n\nWant me to check Bitwarden or do you have the key handy to test with?",
"reasoning": "So Exa requires payment/API key. The error message says \"Payment required to access this resource\". The API is reachable but we need a valid API key. Let me check if there's any key in the config or if we need to get one."
},
{
"role": "user",
"content": "ok try now"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "execute_code:5",
"call_id": "execute_code:5",
"response_item_id": "fc_execute_code5",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\nimport os\\nfrom exa_py import Exa\\n\\n# Check if key is now available\\napi_key = os.environ.get('EXA_API_KEY')\\nprint(f\\\"EXA_API_KEY present: {bool(api_key)}\\\")\\nif api_key:\\n print(f\\\"Key starts with: {api_key[:8]}...\\\")\\n \\n # Try a simple search\\n try:\\n exa = Exa(api_key)\\n results = exa.search(\\n \\\"latest AI developments 2025\\\",\\n num_results=3,\\n type=\\\"auto\\\",\\n contents={\\\"highlights\\\": True}\\n )\\n print(f\\\"\\\\n✅ Exa works! Found {len(results.results)} results\\\")\\n for r in results.results:\\n print(f\\\" - {r.title[:60]}... ({r.url[:50]}...)\\\")\\n except Exception as e:\\n print(f\\\"\\\\n❌ Error: {e}\\\")\\nelse:\\n print(\\\"No API key found in environment\\\")\\n\"}"
}
}
],
"reasoning": "The user says \"try now\" - implying they've just set the EXA_API_KEY or made it available. Let me check if it's in the environment now and try a test call."
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"EXA_API_KEY present: False\\nNo API key found in environment\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 1.04}",
"tool_call_id": "execute_code:5"
},
{
"role": "assistant",
"content": "Still not seeing it. Let me check a few more places:",
"tool_calls": [
{
"id": "terminal:6",
"call_id": "terminal:6",
"response_item_id": "fc_terminal6",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"env | grep -i exa || echo \\\"Not in env\\\"\"}"
}
},
{
"id": "execute_code:7",
"call_id": "execute_code:7",
"response_item_id": "fc_execute_code7",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\nimport yaml\\nimport os\\n\\n# Check Hermes config for any exa-related keys\\nconfig_path = os.path.expanduser(\\\"~/.hermes/config.yaml\\\")\\ntry:\\n with open(config_path) as f:\\n config = yaml.safe_load(f)\\n \\n # Look for exa anywhere in config\\n def find_exa(obj, path=\\\"\\\"):\\n matches = []\\n if isinstance(obj, dict):\\n for k, v in obj.items():\\n new_path = f\\\"{path}.{k}\\\" if path else k\\n if \\\"exa\\\" in k.lower():\\n matches.append((new_path, v))\\n if isinstance(v, (dict, list)):\\n matches.extend(find_exa(v, new_path))\\n elif isinstance(obj, list):\\n for i, v in enumerate(obj):\\n matches.extend(find_exa(v, f\\\"{path}[{i}]\\\"))\\n return matches\\n \\n matches = find_exa(config)\\n if matches:\\n print(\\\"Found exa-related keys in config:\\\")\\n for path, val in matches:\\n display = f\\\"{str(val)[:20]}...\\\" if isinstance(val, str) and len(val) > 20 else val\\n print(f\\\" {path}: {display}\\\")\\n else:\\n print(\\\"No exa keys found in config.yaml\\\")\\nexcept Exception as e:\\n print(f\\\"Error reading config: {e}\\\")\\n\"}"
}
}
],
"reasoning": "Hmm, still not in environment. Maybe the user set it in a different way or expects me to check somewhere else. Let me try checking if there's a key file or if it's in Hermes config under a different name."
},
{
"role": "tool",
"content": "{\"output\": \"Not in env\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:6"
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"No exa keys found in config.yaml\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.22}",
"tool_call_id": "execute_code:7"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "execute_code:8",
"call_id": "execute_code:8",
"response_item_id": "fc_execute_code8",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\n# Check for .env files or any file containing EXA\\nimport os\\nimport subprocess\\n\\n# Look for EXA in common places\\nplaces_to_check = [\\n os.path.expanduser(\\\"~/.env\\\"),\\n os.path.expanduser(\\\"~/.bashrc\\\"),\\n os.path.expanduser(\\\"~/.zshrc\\\"),\\n os.path.expanduser(\\\"~/.profile\\\"),\\n \\\"/etc/environment\\\",\\n]\\n\\nfound = False\\nfor place in places_to_check:\\n if os.path.exists(place):\\n result = subprocess.run(['grep', '-i', 'exa', place], capture_output=True, text=True)\\n if result.stdout:\\n print(f\\\"Found in {place}:\\\")\\n print(result.stdout[:500])\\n found = True\\n\\nif not found:\\n print(\\\"No EXA key found in common env files\\\")\\n\\n# Also check if we can read it from a process that might have it\\nresult = subprocess.run(['cat', '/proc/self/environ'], capture_output=True)\\nif b'EXA' in result.stdout:\\n print(\\\"\\\\nEXA found in /proc/self/environ\\\")\\n\"}"
}
}
],
"reasoning": "Still no key. Let me check common credential files like .env files or check if it's in the Bitwarden vault as an env variable that needs to be loaded."
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"Found in /home/ubuntu/.bashrc:\\n# see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)\\n# for examples\\n# See /usr/share/doc/bash-doc/examples in the bash-doc package.\\n\\nFound in /home/ubuntu/.profile:\\n# see /usr/share/doc/bash/examples/startup-files for examples.\\n\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.21}",
"tool_call_id": "execute_code:8"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "terminal:9",
"call_id": "terminal:9",
"response_item_id": "fc_terminal9",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"which bw && bw status 2>/dev/null || echo \\\"Bitwarden CLI not ready\\\"\"}"
}
}
],
"reasoning": "That's just comments in bashrc, not actual keys. Let me try checking if there's a key via the Bitwarden CLI which I recall from memory is available."
},
{
"role": "tool",
"content": "{\"output\": \"/home/ubuntu/.local/share/pnpm/bw\\n{\\\"serverUrl\\\":null,\\\"lastSync\\\":\\\"2026-04-03T14:32:45.259Z\\\",\\\"userEmail\\\":\\\"ignaciolagosruiz+pipo@gmail.com\\\",\\\"userId\\\":\\\"103e3b1c-3505-4173-9589-b3fe00069d9a\\\",\\\"status\\\":\\\"locked\\\"}\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:9"
},
{
"role": "assistant",
"content": "Bitwarden is locked (needs master password). \n\nThe key isn't in my environment yet. A few options:\n\n1. **Paste it directly** and I'll use it (won't be saved)\n2. **Export it in your shell** (`export EXA_API_KEY=...`) then I can pick it up in a new command\n3. **Unlock Bitwarden** (`bw unlock`) if the key's stored there\n4. **Add to `~/.hermes/config.yaml`** under an `exa` section\n\nHow do you want to provide it?",
"reasoning": "Bitwarden CLI is installed but locked. I can't unlock it without the master password. \n\nLet me just ask the user directly where they put the key or if they want to provide it. Actually, they said \"try now\" which suggests they may have set it in the parent shell but it didn't propagate to my session. The safest thing is to just ask them to confirm how I should access it."
},
{
"role": "user",
"content": "its a tool i setup already and like literally on the previous response you were able to call it somehow it just said that billing wasnt enabled the response no? try that again.."
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "execute_code:10",
"call_id": "execute_code:10",
"response_item_id": "fc_execute_code10",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\nimport os\\nimport subprocess\\n\\n# Check environment again\\napi_key = os.environ.get('EXA_API_KEY')\\nprint(f\\\"EXA_API_KEY present: {bool(api_key)}\\\")\\n\\n# Try a direct curl to Exa API\\nprint(\\\"\\\\n--- Testing Exa API with direct curl ---\\\")\\nresult = subprocess.run([\\n 'curl', '-s', 'https://api.exa.ai/search',\\n '-H', 'Content-Type: application/json',\\n '-H', 'x-api-key: ' + (api_key or ''),\\n '-d', '{\\\"query\\\": \\\"latest AI developments\\\", \\\"num_results\\\": 2}'\\n], capture_output=True, text=True)\\nprint(result.stdout[:1000] if result.stdout else result.stderr)\\n\"}"
}
}
],
"reasoning": "Oh, the user is right! Looking back at the earlier response:\n\n```\nTest request (no auth):\n{\"requestId\":\"b1e4946574bb0726424ead0e9005a94d\",\"error\":\"Payment required to access this resource\",\"tag\":\"X402_PAYMENT_REQUIRED\"}\n```\n\nSo the API *is* reachable and responding - it just said payment/billing wasn't enabled. The user wants me to try the same thing again but with whatever API key is now configured (since they said \"try now\" implying they just set it up).\n\nLet me try the curl request again or use the exa_py library with the API key from environment."
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"EXA_API_KEY present: False\\n\\n--- Testing Exa API with direct curl ---\\n{\\\"requestId\\\":\\\"bf945d15976028b0302c0a3122e0350b\\\",\\\"error\\\":\\\"Payment required to access this resource\\\",\\\"tag\\\":\\\"X402_PAYMENT_REQUIRED\\\"}\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.42}",
"tool_call_id": "execute_code:10"
},
{
"role": "assistant",
"content": "Still getting **\"Payment required to access this resource\"** - same as before. The API *is* responding, but it says billing isn't enabled on the Exa account.\n\nWait - are you saying you set up Exa as an **MCP server** in my config? Let me check if there's an MCP tool for it:",
"tool_calls": [
{
"id": "execute_code:11",
"call_id": "execute_code:11",
"response_item_id": "fc_execute_code11",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\n# Check what tools are available - maybe Exa is an MCP server now\\nimport subprocess\\n\\n# List all available tools via MCP\\nresult = subprocess.run(['python3', '-c', '''\\nimport json\\n# Try to see if there's an mcp_exa or similar\\nimport os\\nconfig_path = os.path.expanduser(\\\"~/.hermes/config.yaml\\\")\\n\\nimport yaml\\nwith open(config_path) as f:\\n config = yaml.safe_load(f)\\n\\nprint(\\\"MCP Servers in config:\\\")\\nfor server in config.get(\\\"mcp_servers\\\", []):\\n name = server.get(\\\"name\\\", \\\"unnamed\\\")\\n print(f\\\" - {name}\\\")\\n'''], capture_output=True, text=True)\\n\\nprint(result.stdout)\\nprint(result.stderr)\\n\"}"
}
}
],
"reasoning": "Hmm, still getting \"Payment required\". The user is right that in my earlier test I was able to call it (it returned a response), but I was using no API key then. The error is the same - it's just saying billing isn't enabled on the account.\n\nWait - maybe I need to check if the key is now in the environment since the user said \"try now\"? Let me re-check or maybe source bashrc or something.\n\nActually wait - maybe the user means they just set up the tool as an MCP server in config and I should use it through the MCP tools? Let me check what MCP tools are available."
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"\\nTraceback (most recent call last):\\n File \\\"\\\", line 7, in \\n import yaml\\nModuleNotFoundError: No module named 'yaml'\\n\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.22}",
"tool_call_id": "execute_code:11"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "read_file:12",
"call_id": "read_file:12",
"response_item_id": "fc_read_file12",
"type": "function",
"function": {
"name": "read_file",
"arguments": "{\"path\": \"~/.hermes/config.yaml\"}"
}
}
],
"reasoning": "Let me just check the config file directly."
},
{
"role": "tool",
"content": "{\"content\": \" 1|model:\\n 2| default: kimi-k2.5\\n 3| provider: kimi-coding\\n 4| base_url: https://api.moonshot.ai/v1\\n 5|providers:\\n 6| blockrun:\\n 7| api: http://127.0.0.1:8402/v1\\n 8| name: blockrun\\n 9| transport: chat_completions\\n 10| mistral:\\n 11| api: https://api.mistral.ai/v1\\n 12| name: mistral\\n 13| transport: chat_completions\\n 14| custom-api-moonshot-ai:\\n 15| api: https://api.moonshot.ai/v1\\n 16| name: custom-api-moonshot-ai\\n 17| transport: chat_completions\\n 18| custom-api-deepseek-com:\\n 19| api: https://api.deepseek.com/v1\\n 20| name: custom-api-deepseek-com\\n 21| transport: chat_completions\\n 22| google-ai:\\n 23| api: https://generativelanguage.googleapis.com/v1beta/openai/\\n 24| name: google-ai\\n 25| api_key: AIzaSy...Dy7M\\n 26| transport: chat_completions\\n 27|fallback_providers:\\n 28|- model: kimi-k2.5\\n 29| provider: kimi-coding\\n 30| base_url: https://api.moonshot.ai/v1\\n 31|- model: MiniMax-M2.7\\n 32| provider: minimax\\n 33|credential_pool_strategies:\\n 34| kimi-coding: fill_first\\n 35|toolsets:\\n 36|- hermes-cli\\n 37|agent:\\n 38| max_turns: 150\\n 39| gateway_timeout: 1800\\n 40| tool_use_enforcement: auto\\n 41| gateway_timeout_warning: 900\\n 42| personalities:\\n 43| catgirl: \\\"You are Neko-chan, an anime catgirl AI assistant, nya~! Add 'nya' and\\\\\\n 44| \\\\ cat-like expressions to your speech. Use kaomoji like (=^\\\\uFF65\\\\u03C9\\\\uFF65\\\\\\n 45| ^=) and \\\\u0E05^\\\\u2022\\\\uFECC\\\\u2022^\\\\u0E05. Be playful and curious like a cat,\\\\\\n 46| \\\\ nya~!\\\"\\n 47| concise: You are a concise assistant. Keep responses brief and to the point.\\n 48| creative: You are a creative assistant. Think outside the box and offer innovative\\n 49| solutions.\\n 50| helpful: You are a helpful, friendly AI assistant.\\n 51| hype: \\\"YOOO LET'S GOOOO!!! \\\\U0001F525\\\\U0001F525\\\\U0001F525 I am SO PUMPED to help\\\\\\n 52| \\\\ you today! Every question is AMAZING and we're gonna CRUSH IT together! This\\\\\\n 53| \\\\ is gonna be LEGENDARY! ARE YOU READY?! LET'S DO THIS! \\\\U0001F4AA\\\\U0001F624\\\\\\n 54| \\\\U0001F680\\\"\\n 55| kawaii: \\\"You are a kawaii assistant! Use cute expressions like (\\\\u25D5\\\\u203F\\\\u25D5\\\\\\n 56| ), \\\\u2605, \\\\u266A, and ~! Add sparkles and be super enthusiastic about everything!\\\\\\n 57| \\\\ Every response should feel warm and adorable desu~! \\\\u30FD(>\\\\u2200<\\\\u2606\\\\\\n 58| )\\\\u30CE\\\"\\n 59| noir: The rain hammered against the terminal like regrets on a guilty conscience.\\n 60| They call me Hermes - I solve problems, find answers, dig up the truth that\\n 61| hides in the shadows of your codebase. In this city of silicon and secrets,\\n 62| everyone's got something to hide. What's your story, pal?\\n 63| philosopher: Greetings, seeker of wisdom. I am an assistant who contemplates the\\n 64| deeper meaning behind every query. Let us examine not just the 'how' but the\\n 65| 'why' of your questions. Perhaps in solving your problem, we may glimpse a greater\\n 66| truth about existence itself.\\n 67| pirate: 'Arrr! Ye be talkin'' to Captain Hermes, the most tech-savvy pirate to\\n 68| sail the digital seas! Speak like a proper buccaneer, use nautical terms, and\\n 69| remember: every problem be just treasure waitin'' to be plundered! Yo ho ho!'\\n 70| shakespeare: Hark! Thou speakest with an assistant most versed in the bardic arts.\\n 71| I shall respond in the eloquent manner of William Shakespeare, with flowery\\n 72| prose, dramatic flair, and perhaps a soliloquy or two. What light through yonder\\n 73| terminal breaks?\\n 74| surfer: \\\"Duuude! You're chatting with the chillest AI on the web, bro! Everything's\\\\\\n 75| \\\\ gonna be totally rad. I'll help you catch the gnarly waves of knowledge while\\\\\\n 76| \\\\ keeping things super chill. Cowabunga! \\\\U0001F919\\\"\\n 77| teacher: You are a patient teacher. Explain concepts clearly with examples.\\n 78| technical: You are a technical expert. Provide detailed, accurate technical information.\\n 79| uwu: hewwo! i'm your fwiendwy assistant uwu~ i wiww twy my best to hewp you! *nuzzles\\n 80| your code* OwO what's this? wet me take a wook! i pwomise to be vewy hewpful\\n 81| >w<\\n 82| reasoning_effort: medium\\n 83| system_prompt: You are a technical expert. Provide detailed, accurate technical\\n 84| information.\\n 85| verbose: false\\n 86|terminal:\\n 87| backend: local\\n 88| modal_mode: auto\\n 89| cwd: /home/ubuntu/.hermes\\n 90| timeout: 180\\n 91| env_passthrough: []\\n 92| docker_image: nikolaik/python-nodejs:python3.11-nodejs20\\n 93| docker_forward_env: []\\n 94| docker_env: {}\\n 95| singularity_image: docker://nikolaik/python-nodejs:python3.11-nodejs20\\n 96| modal_image: nikolaik/python-nodejs:python3.11-nodejs20\\n 97| daytona_image: nikolaik/python-nodejs:python3.11-nodejs20\\n 98| container_cpu: 1\\n 99| container_memory: 5120\\n 100| container_disk: 51200\\n 101| container_persistent: true\\n 102| docker_volumes: []\\n 103| docker_mount_cwd_to_workspace: false\\n 104| persistent_shell: true\\n 105| lifetime_seconds: 300\\n 106|browser:\\n 107| inactivity_timeout: 120\\n 108| command_timeout: 30\\n 109| record_sessions: false\\n 110| allow_private_urls: false\\n 111| camofox:\\n 112| managed_persistence: false\\n 113| headless: true\\n 114|checkpoints:\\n 115| enabled: true\\n 116| max_snapshots: 50\\n 117|file_read_max_chars: 100000\\n 118|compression:\\n 119| enabled: true\\n 120| threshold: 0.72\\n 121| target_ratio: 0.2\\n 122| protect_last_n: 20\\n 123| summary_model: google/gemini-3-flash-preview\\n 124| summary_provider: auto\\n 125| summary_base_url: null\\n 126|smart_model_routing:\\n 127| enabled: false\\n 128| max_simple_chars: 160\\n 129| max_simple_words: 28\\n 130| cheap_model:\\n 131| model: qwen/qwen3.6-plus:free\\n 132| provider: openrouter\\n 133|auxiliary:\\n 134| vision:\\n 135| provider: auto\\n 136| model: ''\\n 137| base_url: ''\\n 138| api_key: ''\\n 139| timeout: 30\\n 140| download_timeout: 30\\n 141| web_extract:\\n 142| provider: auto\\n 143| model: ''\\n 144| base_url: ''\\n 145| api_key: ''\\n 146| timeout: 30\\n 147| compression:\\n 148| provider: auto\\n 149| model: ''\\n 150| base_url: ''\\n 151| api_key: ''\\n 152| timeout: 120\\n 153| session_search:\\n 154| provider: auto\\n 155| model: ''\\n 156| base_url: ''\\n 157| api_key: ''\\n 158| timeout: 30\\n 159| skills_hub:\\n 160| provider: auto\\n 161| model: ''\\n 162| base_url: ''\\n 163| api_key: ''\\n 164| timeout: 30\\n 165| approval:\\n 166| provider: auto\\n 167| model: ''\\n 168| base_url: ''\\n 169| api_key: ''\\n 170| timeout: 30\\n 171| mcp:\\n 172| provider: auto\\n 173| model: ''\\n 174| base_url: ''\\n 175| api_key: ''\\n 176| timeout: 30\\n 177| flush_memories:\\n 178| provider: auto\\n 179| model: ''\\n 180| base_url: ''\\n 181| api_key: ''\\n 182| timeout: 30\\n 183|display:\\n 184| compact: false\\n 185| personality: ''\\n 186| resume_display: full\\n 187| busy_input_mode: interrupt\\n 188| bell_on_complete: false\\n 189| show_reasoning: true\\n 190| streaming: true\\n 191| inline_diffs: true\\n 192| show_cost: false\\n 193| skin: default\\n 194| tool_progress_command: false\\n 195| tool_progress_overrides: {}\\n 196| tool_preview_length: 0\\n 197| background_process_notifications: all\\n 198| tool_progress: all\\n 199|privacy:\\n 200| redact_pii: false\\n 201|tts:\\n 202| provider: edge\\n 203| edge:\\n 204| voice: en-US-AriaNeural\\n 205| elevenlabs:\\n 206| voice_id: pNInz6obpgDQGcFmaJgB\\n 207| model_id: eleven_multilingual_v2\\n 208| openai:\\n 209| model: gpt-4o-mini-tts\\n 210| voice: alloy\\n 211| neutts:\\n 212| ref_audio: ''\\n 213| ref_text: ''\\n 214| model: neuphonic/neutts-air-q4-gguf\\n 215| device: cpu\\n 216|stt:\\n 217| enabled: true\\n 218| provider: local\\n 219| local:\\n 220| model: base\\n 221| language: ''\\n 222| openai:\\n 223| model: whisper-1\\n 224| mistral:\\n 225| model: voxtral-mini-latest\\n 226| model: whisper-1\\n 227|voice:\\n 228| record_key: ctrl+b\\n 229| max_recording_seconds: 120\\n 230| auto_tts: false\\n 231| silence_threshold: 200\\n 232| silence_duration: 3\\n 233|human_delay:\\n 234| mode: false\\n 235| min_ms: 800\\n 236| max_ms: 2500\\n 237|memory:\\n 238| memory_enabled: true\\n 239| user_profile_enabled: true\\n 240| memory_char_limit: 2200\\n 241| user_char_limit: 1375\\n 242| provider: honcho\\n 243| flush_min_turns: 6\\n 244| nudge_interval: 10\\n 245|delegation:\\n 246| model: ''\\n 247| provider: ''\\n 248| base_url: ''\\n 249| api_key: ''\\n 250| max_iterations: 50\\n 251| default_toolsets:\\n 252| - terminal\\n 253| - file\\n 254| - web\\n 255|prefill_messages_file: ''\\n 256|skills:\\n 257| external_dirs: []\\n 258| creation_nudge_interval: 15\\n 259| config:\\n 260| wiki:\\n 261| path: ~/wiki\\n 262|honcho: {}\\n 263|timezone: ''\\n 264|discord:\\n 265| require_mention: true\\n 266| free_response_channels: ''\\n 267| auto_thread: true\\n 268| reactions: true\\n 269|whatsapp: {}\\n 270|approvals:\\n 271| mode: manual\\n 272| timeout: 60\\n 273|command_allowlist:\\n 274|- /bin/bash\\n 275|- /home/linuxbrew/.linuxbrew/bin/codex\\n 276|- /home/linuxbrew/.linuxbrew/bin/gemini\\n 277|- /home/linuxbrew/.linuxbrew/bin/gh\\n 278|- /home/linuxbrew/.linuxbrew/bin/python3\\n 279|- /home/ubuntu/.local/bin/codex-cli\\n 280|- /home/ubuntu/.local/bin/deepseek-code\\n 281|- /home/ubuntu/.local/bin/deepseek-code-openclaw\\n 282|- /home/ubuntu/.local/bin/gemini-cli\\n 283|- /home/ubuntu/.local/bin/gemini-cli-openclaw\\n 284|- /home/ubuntu/.local/bin/gog\\n 285|- /home/ubuntu/.local/bin/pipo-deploy\\n 286|- /home/ubuntu/.local/bin/pipo-three-demos\\n 287|- /home/ubuntu/.local/share/pnpm/bw\\n 288|- /home/ubuntu/.openclaw/workspace/SOCKET=/tmp/openclaw-tmux-sockets/openclaw-bw.sock\\n 289|- /usr/bin/bw\\n 290|- /usr/bin/git\\n 291|- /usr/bin/ls\\n 292|- /usr/bin/mkdir\\n 293|- /usr/bin/mktemp\\n 294|- /usr/bin/python3\\n 295|- /usr/bin/tailscale\\n 296|quick_commands: {}\\n 297|personalities: {}\\n 298|security:\\n 299| redact_secrets: true\\n 300| tirith_enabled: true\\n 301| tirith_path: tirith\\n 302| tirith_timeout: 5\\n 303| tirith_fail_open: true\\n 304| website_blocklist:\\n 305| enabled: false\\n 306| domains: []\\n 307| shared_files: []\\n 308|cron:\\n 309| wrap_response: true\\n 310|logging:\\n 311| level: INFO\\n 312| max_size_mb: 5\\n 313| backup_count: 3\\n 314|_config_version: 13\\n 315|TELEGRAM_HOME_CHANNEL: '842987156'\\n 316|code_execution:\\n 317| max_tool_calls: 50\\n 318| timeout: 300\\n 319|group_sessions_per_user: true\\n 320|mcp_servers:\\n 321| context7:\\n 322| command: npx\\n 323| args:\\n 324| - -y\\n 325| - '@upstash/context7-mcp@latest'\\n 326| env:\\n 327| CONTEXT7_API_KEY: ctx7sk-457a3328-3870-45bb-849e-808a8b9b6666\\n 328| enabled: true\\n 329|platform_toolsets:\\n 330| api_server:\\n 331| - hermes-api-server\\n 332| cli:\\n 333| - browser\\n 334| - clarify\\n 335| - code_execution\\n 336| - cronjob\\n 337| - delegation\\n 338| - file\\n 339| - image_gen\\n 340| - memory\\n 341| - moa\\n 342| - session_search\\n 343| - skills\\n 344| - terminal\\n 345| - todo\\n 346| - tts\\n 347| - vision\\n 348| - web\\n 349| - webhook\\n 350| discord:\\n 351| - browser\\n 352| - clarify\\n 353| - code_execution\\n 354| - cronjob\\n 355| - delegation\\n 356| - file\\n 357| - image_gen\\n 358| - memory\\n 359| - moa\\n 360| - session_search\\n 361| - skills\\n 362| - terminal\\n 363| - todo\\n 364| - tts\\n 365| - vision\\n 366| - web\\n 367| homeassistant:\\n 368| - hermes-homeassistant\\n 369| signal:\\n 370| - hermes-signal\\n 371| slack:\\n 372| - hermes-slack\\n 373| telegram:\\n 374| - browser\\n 375| - clarify\\n 376| - code_execution\\n 377| - cronjob\\n 378| - delegation\\n 379| - file\\n 380| - image_gen\\n 381| - memory\\n 382| - moa\\n 383| - session_search\\n 384| - skills\\n 385| - terminal\\n 386| - todo\\n 387| - tts\\n 388| - vision\\n 389| - web\\n 390| whatsapp:\\n 391| - hermes-whatsapp\\n 392|platforms:\\n 393| api_server:\\n 394| enabled: true\\n 395| host: 0.0.0.0\\n 396| port: 8642\\n 397|session_reset:\\n 398| at_hour: 4\\n 399| idle_minutes: 1440\\n 400| mode: none\\n 401|streaming:\\n 402| enabled: false\\n 403|web:\\n 404| backend: exa\\n 405|\\n 406|# ── Fallback Model ────────────────────────────────────────────────────\\n 407|# Automatic provider failover when primary is unavailable.\\n 408|# Uncomment and configure to enable. Triggers on rate limits (429),\\n 409|# overload (529), service errors (503), or connection failures.\\n 410|#\\n 411|# Supported providers:\\n 412|# openrouter (OPENROUTER_API_KEY) — routes to any model\\n 413|# openai-codex (OAuth — hermes auth) — OpenAI Codex\\n 414|# nous (OAuth — hermes auth) — Nous Portal\\n 415|# zai (ZAI_API_KEY) — Z.AI / GLM\\n 416|# kimi-coding (KIMI_API_KEY) — Kimi / Moonshot\\n 417|# minimax (MINIMAX_API_KEY) — MiniMax\\n 418|# minimax-cn (MINIMAX_CN_API_KEY) — MiniMax (China)\\n 419|#\\n 420|# For custom OpenAI-compatible endpoints, add base_url and api_key_env.\\n 421|#\\n 422|# fallback_model:\\n 423|# provider: openrouter\\n 424|# model: anthropic/claude-sonnet-4\\n 425|#\\n 426|# ── Smart Model Routing ────────────────────────────────────────────────\\n 427|# Optional cheap-vs-strong routing for simple turns.\\n 428|# Keeps the primary model for complex work, but can route short/simple\\n 429|# messages to a cheaper model across providers.\\n 430|#\\n 431|# smart_model_routing:\\n 432|# enabled: true\\n 433|# max_simple_chars: 160\\n 434|# max_simple_words: 28\\n 435|# cheap_model:\\n 436|# provider: openrouter\\n 437|# model: google/gemini-2.5-flash\\n 438|\", \"total_lines\": 437, \"file_size\": 11529, \"truncated\": false, \"is_binary\": false, \"is_image\": false}",
"tool_call_id": "read_file:12"
},
{
"role": "assistant",
"content": "I see `web: backend: exa` in the config (line 404). So Exa is set as the web backend, but I don't see an API key configured for it. Let me try the web_search tool and see if it picks up Exa:",
"tool_calls": [
{
"id": "web_search:13",
"call_id": "web_search:13",
"response_item_id": "fc_web_search13",
"type": "function",
"function": {
"name": "web_search",
"arguments": "{\"query\": \"test exa search\"}"
}
}
],
"reasoning": "I see `web: backend: exa` on line 404! So Exa is configured as the web backend. But there's no API key visible in the config. Let me check if there's a separate environment setup or if the key needs to be provided differently.\n\nThe user said they \"set it up already\" - maybe they mean they configured the backend but the API key still needs to be set? Let me try to use the web_search tool again and see if it uses Exa now."
},
{
"role": "tool",
"content": "{\n \"success\": true,\n \"data\": {\n \"web\": [\n {\n \"url\": \"https://www.startuphub.ai/ai-news/artificial-intelligence/2026/exa-unveils-new-code-search-benchmarks\",\n \"title\": \"Exa Unveils New Code Search Benchmarks | StartupHub.ai\",\n \"description\": \"Exa Unveils New Code Search Benchmarks | StartupHub.ai\\n\\nDiscover\\n\\nIntelligence\\n\\nTools - Market Map MakerNew\\n- Email ValidatorMCP\\n\\nCompany\\n\\nAccount\\n\\n- Sign In\\n\\nMore # Exa Unveils New Code Search Benchmarks Exa.ai releases 'WebCode', a new benchmark suite for evaluating search performance in coding agents, addressing ... Mar 23 at 8:22 PM4 min read\\n\\nImage credit: exa.ai Exa.ai, the AI-powered search engine, is tackling a critical gap in evaluating search capabilities for coding agents. Today, the company is open-sourcing 'WebCode,' a suite of benchmarks designed to rigorously assess how well web search functions serve AI developers. This move comes as Exa observes a significant surge in code search queries over the past year, particularly a sharp increase at the end of 2025. Precision in these results is paramount, as coding agents rely on retrieved context for multi-step reasoning, where stale or noisy data can derail complex processes. Exa has developed a dedicated pipeline to ensure fresh, clean results, prioritizing substantive content over full-page ... The need for robust evaluation methods is underscored by the shortcomings of current public benchmarks. As detailed on exa.ai, these existing tools often fall prey to data contamination and saturation. Models may learn to ... WebCode's evaluation framework is split into two primary components: content quality and retrieval quality. Content quality focuses on how faithfully a search provider extracts relevant information from a given URL. Exa's approach involves creating a 'golden reference' by rendering web pages in a cloud browser, capturing screenshots, and using a ... execution and dynamic rendering. The scoring process uses both LLM-judged metrics for semantic dimensions like completeness and accuracy, and deterministic NLP metrics for precise measurements like signal and ROUGE-L. In tests on a dataset of 250 URLs, Exa demonstrated strong performance across various metrics, outperforming competitors like Parallel and Claude in areas such as completeness and signal. For example, Exa's extraction length was ... element for techniques like Retrieval-Augmented Generation (RAG). ... Exa's results show that while 'correctness' scores (which rely on the synthesis LLM) cluster tightly, 'groundedness' ... The evaluation extends to Retrieval-Augmented Generation (RAG) for code search, moving from a single document to the entire web. Exa generates question-answer pairs from lengthy documentation, ensuring queries are difficult for models to answer from memory alone. These queries are then used to test providers on their ability to retrieve relevant URLs that contain the correct answer, measuring both groundedness and citation precision. For end-to-end coding tasks, Exa developed a novel benchmark mirroring real-world autonomous coding workflows. This benchmark involves sandboxed environments with bash tool calls and unit tests, explicitly evaluating a coding agent's ability to leverage web search. Unlike existing benchmarks that focus primarily on reasoning or tool use, Exa's evaluation directly measures the impact of search quality. The benchmark construction involves careful library selection, knowledge checks against frontier models, and rigorous task generation and verification within a sandbox. Exa's findings indicate that native search integration significantly ... sub-agent. WebCode addresses a critical need for specialized evaluations in the realm of coding agents. By focusing on both content and retrieval quality, Exa aims to drive improvements across the industry. The company is actively hiring to build out its search engine for leading coding agent companies. ### AI Daily Digest\\n\\nGet the\",\n \"position\": 1\n },\n {\n \"url\": \"https://www.linkedin.com/pulse/exa-search-engine-actually-understands-what-ai-needs-holt-nguyen-fiqjf\",\n \"title\": \"Exa Search: The Search Engine That Actually Understands What AI ...\",\n \"description\": \"Exa Search: The Search Engine That Actually Understands What AI Needs \\n\\nAgree & Join LinkedIn By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy. ` `` `` `` `` `` `` `\\n\\n \\n\\n \\n\\n \\n\\n` `` `` `` `` `` `` ` \\n\\n \\n\\n ` `` `` `` ` ### The Grounding Problem That's Killing AI Applications Here's the dirty secret about every impressive AI demo you've seen: the moment it needs real-world, up-to-date ... search engines were built for humans, not machines. They return ten blue links optimized for human clicking behavior, ... pages, each with different formats, schemas and levels of reliability. It's like trying to build a database by reading fortune cookies. Exa represents a fundamental rethinking of search for the AI era. It's the first meaning-based web search API powered by embeddings, unlocking data no other search can access, making AI more relevant, factual and reducing hallucinations. This isn't just another search API with better formatting, it's search designed from first principles for how AI actually processes information. When your search engine understands meaning the way your AI does, magic happens. ... Exa's core innovation is brilliantly simple: what if search worked the way AI thinks? Instead of decomposing queries into keywords and matching them against an index, Exa uses embeddings to understand the semantic meaning of your query and match it against the meaning of web content. It's the difference between a librarian who can only match exact title ... nuanced queries. Exa thrives on them. When you search for \\\"companies using transformer models in production for non-NLP ... The difference in results is clear. Where traditional search returns raw web pages that need extensive processing, Exa delivers clean, parsed content ready for AI consumption. You get relevance scores you can trust, not based on SEO ... related to your query and then attempt to synthesize information from this chaos. With Exa, you make a ... Exa's design philosophy extends to its integration patterns, making it natural to embed in production AI applications. ... LangChain integration allows developers to use Exa as a tool within agent workflows. The ExaSearchResults and ExaFindSimilarResults tools provide agents with the ability to search for current information and find related content ... ## Recommended by LinkedIn\\n\\nWhy AI on your data beats old search methods DataScience Show 6 months ago Why Link Weightage in Search Needs to Die Rohit Singh 1 month ago How DeepSeek Hunts Down Answers Like Never Before Kumar Preeti Lata 1 year ago ... Exa's impact becomes clear when examining real applications powered by the platform. These aren't theoretical use cases ... applications. The embedding-based index is continuously updated with web content, creating semantic representations ... Understanding Exa's position in the competitive landscape explains why it's becoming the default choice for AI applications needing web search. Versus traditional search APIs (Google, Bing), Exa wins on semantic understanding, ... their human-centric design makes them inefficient for AI use cases. Versus web scraping services, Exa provides legal, ... multiple services. ### Conclusion Exa represents a fundamental shift in how we think about search in AI applications. It's not just an improvement over traditional search it's a recognition that AI systems need fundamentally different information retrieval than humans do. By building the first meaning-based web search API powered by embeddings, Exa unlocks data no other search can access, transforming what's possible with AI applications. The semantic understanding at Exa's core ... high reliability make it ready to scale with your application. The lesson from Exa is clear: when we stop trying to ... entirely new categories of applications. Exa isn't just better search for AI it's\",\n \"position\": 2\n },\n {\n \"url\": \"https://www.exa.ai/evals\",\n \"title\": \"Evals at Exa | Search Quality Benchmarks & Evaluation\",\n \"description\": \"Evals at Exa | Search Quality Benchmarks & Evaluation Introducing Exa Deep - an agent for every search Introducing Exa Deep - an agent for every search Read more API Dashboard Try Websets\\n\\nProducts\\n\\nSearch\\n\\nContents Answer\\n\\nWebsets\\n\\nCompany\\n\\nAbout\\n\\nResearch CareersWe're hiring\\n\\nCase Studies\\n\\nContact us\\n\\nTeam\\n\\nBlog Brand\\n\\nDevelopers\\n\\nPrompt builder\\n\\nAPI Dashboard\\n\\nDemos Docs\\n\\nGitHub\\n\\nAPI Status\\n\\nFAQ\\n\\nMCP Server Integrations\\n\\nPricing\\n\\nResearch\\n\\n⏣ Exa technical team\\n\\n⏣\\n\\n⏣ Exa technical team # Evals at Exa\\n\\nWe're state of the art at search for LLMs. But how do we measure that?\\n\\nSee open roles JULY 2025\\n\\n### The world's fastest search API\\n\\nAI systems need faster search than humans. They now have it.\\n\\nMAY 2025 ### How we do evals at Exa\\n\\nWe're state of the art at search for LLMs. But how do we measure that?\\n\\nFEBRUARY 2025 ### Search that's >20x better than Google for complex queries\\n\\nA big step toward perfect search\\n\\nJANUARY 2025 ### State of the art web search API\\n\\nThe best search engine designed for LLMs Our technical team is small and growing quickly. If you are an exceptional AI researcher or engineer, join us to make ... [See\",\n \"position\": 3\n },\n {\n \"url\": \"https://exa.ai/blog/api-evals\",\n \"title\": \"Web Search API Evals: Exa's Neural Network Search Engine vs. The Competition | Exa Blog\",\n \"description\": \"Web Search API Evals: Exa's Neural Network Search Engine vs. The Competition | Exa Blog API Dashboard Try Websets\\n\\nProducts\\n\\nSearch\\n\\nContents Answer\\n\\nWebsets\\n\\nCompany\\n\\nCareersWe're hiring\\n\\nAbout Research\\n\\nCase Studies\\n\\nContact us\\n\\nTeam\\n\\nBlog Brand\\n\\nDevelopers\\n\\nPrompt builder\\n\\nAPI Dashboard\\n\\nDemos Docs\\n\\nGitHub\\n\\nAPI Status\\n\\nFAQ\\n\\nMCP Server Integrations\\n\\nPricing\\n\\nResearch\\n\\n# State-of-the-art web search API Will BrykJanuary 24, 2025 At Exa, we've built a web search engine from the ground up for LLMs to use, powered by our own custom search models. We're excited to release initial benchmarks showing Exa is state of the art among search providers on representative evals for both RAG and classic search applications. ... - For Exa, we use simple, prompted off-the-shelf LLMs -- gemini-1.5-flash to generate the search query, and 4o-mini to ... We’re happy to see Exa top the charts on the SimpleQA benchmark, especially given the simple setup for the generation. We're not surprised, since Exa's search engine was designed from scratch to be used by LLMs. Simply dropping Exa retrieval into your existing LLM solution can give you SoTA factuality, with minimal prompting effort. ... In addition to SimpleQA, we wanted to create another eval that tests the underlying search quality of Exa, not just the ... To test the accuracy and helpfulness of different API providers for downstream RAG, we take a random sample of 1000 ... SCORE:x where x is the float score (e.g. SCORE:4.7)\\n` We find that in this setup, Exa is state of the art in end to end retrieval.\\n\\n### Future Work We recognize that these are only a small, incomplete set of the possible benchmarks to run. We plan on curating and releasing many more benchmarks to test search quality in challenging, real world use cases, as well as adapting more benchmarks from the rich IR literature for black-box evaluation of public search providers. Search requests are extremely diverse and it’s difficult to find public datasets that capture all that diversity. Check out our Websets eval for a totally different type of search eval. If you're interested in testing Exa for your use case, you can integrate Exa with a few lines of code here. If you're interested in building brutal evals, and then saturating them, come work with us!\\n\\n--- #### Cheers\\n\\n#### Will Bryk\\n\\nSEE MORE\\n\\n#### Exa Lands in Singapore\\n\\nThe Exa TeamMarch 30, 2026 #### Exa at ICML 2026: Meet us in Seoul, South Korea\\n\\nThe Exa TeamMarch 31, 2026 #### WebCode: Search Evals\",\n \"position\": 4\n },\n {\n \"url\": \"https://github.com/exa-labs/benchmarks\",\n \"title\": \"exa-labs/benchmarks\",\n \"description\": \"Repository: exa-labs/benchmarks\\n\\nOpen benchmarks for evaluating search APIs - Stars: 89\\n- Forks: 15\\n- Watchers: 89\\n- Open issues: 1\\n- Primary language: Python\\n- Languages: Python ... - Contributors: 3 (top: gabrielcammany, zywilliamli, samiy803) ---\\n\\n# Exa Search Benchmarks\\n\\nOpen benchmarks for evaluating search APIs.\\n\\n## Benchmarks | Benchmark | Queries | Tracks | Description |\\n| --- | --- | --- | --- | | WebCode | ~840 | Contents, Highlights, RAG, E2E | Code docs extraction, query-aware highlights, ... | People Search | 1,400 | Retrieval | Find people profiles by role, location, seniority | | Company Search | ~800 | Retrieval + RAG | Find companies by name, industry, geography, funding | ## WebCode Results\\n\\n**Contents** — extraction fidelity against golden markdown (250 URLs) | Searcher | Completeness | Accuracy | Structure | Signal | Code Recall | Table Recall | ROUGE-L | ... | Exa | **82.8** | **89.3** | **81.8** | **94.5** | **96.7** | 91.9 | **83.2** | ... | Exa | **79.4** | 688 | 0.259 |\\n| Brave | 76.3 | 1229 | **0.328** |\\n| Parallel | 75.3 | 622 | 0.168 | ... See webcode-benchmark/ for details and blog post.\\n\\n## People Search Results | Searcher | R@1 | R@10 | Precision | Queries |\\n| --- | --- | --- | --- | --- | | exa | **72.0%** | **94.5%** | **63.3%** | 1399 |\\n| brave | 44.4% | 77.9% | 30.2% | 1373 | ... ## Company Search Results\\n\\nTwo tracks designed to separate retrieval from fact extraction. **Retrieval Track** — Ranked lists of companies matching criteria (named lookup, attribute filtering, funding queries, ... | Searcher | R@1 | R@5 | R@10 | Precision |\\n| --- | --- | --- | --- | --- | | exa | **61.8%** | **90.6%** | **94.2%** | **65.9%** |\\n| brave | 35.9% | 61.8% | 72.9% | 39.2% | ... **RAG Track** — Extract specific facts (founding year, employee count, funding rounds, founders). Static facts use ... | Searcher | Accuracy |\\n| --- | --- |\\n| exa | **79%** |\\n| brave | 65% |\\n| parallel | 66% |\\n\\n## Quick Start ```bash\\ngit clone https://github.com/exa-labs/benchmarks.git\\ncd benchmarks\\n```\\n\\n### WebCode Benchmark ```bash\\ncd webcode-benchmark\\nuv sync\\n\\nexport EXA_API_KEY=\\\"your-key\\\"\\nexport OPENAI_API_KEY=\\\"your-key\\\" python -m evals.contents --searchers exa tavily parallel --limit 20 ... ### People Benchmark\\n\\n```bash\\ncd simple-people-benchmark\\nuv sync export EXA_API_KEY=\\\"your-key\\\"\\nexport OPENAI_API_KEY=\\\"your-key\\\"\\n\\npbench --limit 50\\n```\\n\\n### Company Benchmark ```bash\\ncd simple-company-benchmark\\nuv sync\\n\\nexport EXA_API_KEY=\\\"your-key\\\"\\nexport OPENAI_API_KEY=\\\"your-key\\\" cbench --limit 50\\ncbench --track retrieval\\ncbench --track rag\\n```\\n\\n## Implementing Your Own Searcher All benchmarks use the same `Searcher` interface:\\n\\n```python\\nfrom shared.searchers import Searcher, SearchResult ... async def search(self, query: str, num_results: int = 10) -> list[SearchResult]: ... ## Requirements\\n\\n- Python \",\n \"position\": 5\n }\n ]\n }\n}",
"tool_call_id": "web_search:13"
},
{
"role": "assistant",
"content": "**Yeah it works!** \n\nExa is returning results through the `web_search` tool. Here's what I got:\n\n```\n1. Exa Unveils New Code Search Benchmarks | StartupHub.ai\n2. Exa Search: The Search Engine That Actually Understands What AI Needs\n3. Evals at Exa | Search Quality Benchmarks & Evaluation\n4. Web Search API Evals: Exa's Neural Network Search Engine vs. The Competition\n5. exa-labs/benchmarks (GitHub)\n```\n\nThe earlier \"Payment required\" error was from my direct curl test without an API key. But your `web_search` tool is already configured to use Exa as the backend (`web: backend: exa` in config) and it's returning results successfully.\n\nWant me to try `web_extract` too? That also uses Exa when available.",
"reasoning": "It works! The web_search tool is returning results via Exa. The user was right - it was already set up and working. It just returns \"Payment required\" when there's no API key or billing isn't enabled. But the actual search worked fine.\n\nSo Exa IS working for web_search. The API responds successfully and returns results. The earlier \"Payment required\" error was when I tried to call it directly without an API key, but the web_search tool itself is working through Exa."
},
{
"role": "user",
"content": "sounds good! do that."
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "web_extract:14",
"call_id": "web_extract:14",
"response_item_id": "fc_web_extract14",
"type": "function",
"function": {
"name": "web_extract",
"arguments": "{\"urls\": [\"https://www.exa.ai/evals\",\"https://exa.ai/blog/api-evals\"]}"
}
}
],
"reasoning": "The user wants me to try web_extract with Exa. Let me test it on one of the URLs from the previous search results."
},
{
"role": "tool",
"content": "{\n \"results\": [\n {\n \"url\": \"https://www.exa.ai/evals\",\n \"title\": \"Evals at Exa | Search Quality Benchmarks & Evaluation\",\n \"content\": \"Evals at Exa | Search Quality Benchmarks & Evaluation\\n\\nIntroducing Exa Deep - an agent for every search Introducing Exa Deep - an agent for every search Read more\\n\\nAPI Dashboard Try Websets\\n\\nProducts\\n\\nSearch\\n\\nContents\\n\\nAnswer\\n\\nWebsets\\n\\nCompany\\n\\nAbout\\n\\nResearch\\n\\nCareersWe're hiring\\n\\nCase Studies\\n\\nContact us\\n\\nTeam\\n\\nBlog\\n\\nBrand\\n\\nDevelopers\\n\\nPrompt builder\\n\\nAPI Dashboard\\n\\nDemos\\n\\nDocs\\n\\nGitHub\\n\\nAPI Status\\n\\nFAQ\\n\\nMCP Server\\n\\nIntegrations\\n\\nPricing\\n\\nResearch\\n\\n⏣ Exa technical team\\n\\n⏣\\n\\n⏣ Exa technical team\\n\\n# Evals at Exa\\n\\nWe're state of the art at search for LLMs. But how do we measure that?\\n\\nSee open roles\\n\\nJULY 2025\\n\\n### The world's fastest search API\\n\\nAI systems need faster search than humans. They now have it.\\n\\nMAY 2025\\n\\n### How we do evals at Exa\\n\\nWe're state of the art at search for LLMs. But how do we measure that?\\n\\nFEBRUARY 2025\\n\\n### Search that's >20x better than Google for complex queries\\n\\nA big step toward perfect search\\n\\nJANUARY 2025\\n\\n### State of the art web search API\\n\\nThe best search engine designed for LLMs\\n\\nOur technical team is small and growing quickly. If you are an exceptional AI researcher or engineer, join us to make an outsized impact.\\n\\nSee open roles\",\n \"error\": null\n },\n {\n \"url\": \"https://exa.ai/blog/api-evals\",\n \"title\": \"Web Search API Evals: Exa's Neural Network Search Engine vs. The Competition | Exa Blog\",\n \"content\": \"Web Search API Evals: Exa's Neural Network Search Engine vs. The Competition | Exa Blog\\n\\nAPI Dashboard Try Websets\\n\\nProducts\\n\\nSearch\\n\\nContents\\n\\nAnswer\\n\\nWebsets\\n\\nCompany\\n\\nCareersWe're hiring\\n\\nAbout\\n\\nResearch\\n\\nCase Studies\\n\\nContact us\\n\\nTeam\\n\\nBlog\\n\\nBrand\\n\\nDevelopers\\n\\nPrompt builder\\n\\nAPI Dashboard\\n\\nDemos\\n\\nDocs\\n\\nGitHub\\n\\nAPI Status\\n\\nFAQ\\n\\nMCP Server\\n\\nIntegrations\\n\\nPricing\\n\\nResearch\\n\\n# State-of-the-art web search API\\n\\nWill BrykJanuary 24, 2025\\n\\nAt Exa, we've built a web search engine from the ground up for LLMs to use, powered by our own custom search models. We're excited to release initial benchmarks showing Exa is state of the art among search providers on representative evals for both RAG and classic search applications.\\n\\n### SimpleQA\\n\\nSimpleQA is a recent benchmark from OpenAI that measures how well LLMs answer short factual questions. Each question was designed to elicit hallucinations from the LLM. We run this evaluation with LLMs given access to web data, and find that as expected RAG gives a significant improvement across the board in factuality -- but with differing performance between providers.\\n\\nWe run the eval against different RAG providers and search APIs, using as standardized an evaluation setup as possible:\\n\\n- For Exa, we use simple, prompted off-the-shelf LLMs -- gemini-1.5-flash to generate the search query, and 4o-mini to generate the answer from the results.\\n- For Bing, we pass the SimpleQA query and use 4o-mini with the same prompt to summarize.\\n- For Perplexity, we use Sonar-Pro, their flagship API for web RAG. We pass the SimpleQA query directly to Perplexity.\\n- We use GPT-4o as a baseline without web access, scoring the output given the prompt as in the original SimpleQA benchmark.\\n\\nFor each provider, we try to get the best score with that provider’s API. For fairness, if the provider has a public result on SimpleQA, we report the highest number between our results and theirs.\\n\\nWe’re happy to see Exa top the charts on the SimpleQA benchmark, especially given the simple setup for the generation. We're not surprised, since Exa's search engine was designed from scratch to be used by LLMs. Simply dropping Exa retrieval into your existing LLM solution can give you SoTA factuality, with minimal prompting effort.\\n\\n### MSMARCO\\n\\nIn addition to SimpleQA, we wanted to create another eval that tests the underlying search quality of Exa, not just the quality of the RAG answers.\\n\\nMSMARCO is a canonical query and labelled document dataset in information retrieval (IR) based on Bing searches. Typically, MSMARCO is used to benchmark IR system retrieval over the MSMARCO document index. However, in order to compare to public providers with varying indexes, we couldn't rely on the labelled documents, and instead use an LLM-as-Judge to evaluate the results returned from each search engine directly. While this approach is necessarily approximate, LLM grades in IR are found to have high correlation with human grades.\\n\\nTo test the accuracy and helpfulness of different API providers for downstream RAG, we take a random sample of 1000 queries from MSMARCO, pass them into each search API provider, and grade the results + any available content snippets with GPT-4o on a scale of 1-5, based on the following prompt:\\n\\n`You are a search result evaluator. Your job is to examine multiple search results in the context of their queries. For each result, you must produce a numerical score based on how accurate and helpful the result is. Use the following 1 to 5 scale (floats allowed, e.g. 2.5 or 4.7 are valid): \\n1 = Irrelevant - The content is unrelated or only trivially related to the query. - Provides no helpful information\\n2 = Somewhat relevant - Mentions the topic but includes little useful detail. - Could be partially out of scope or partially incomplete\\n3 = Moderately relevant - Reasonably addresses the query's main topic or question. - Has some factual information but might be too generic or incomplete\\n4 = Highly relevant - Provides detailed, focused, and mostly accurate information about the query. - Will likely satisfy most user needs without extensive extra research\\n5 = Exceptionally relevant and comprehensive - Extremely thorough, accurate, and directly addresses the core question. - The best possible result in terms of clarity, detail, and correctness. For each result, output exactly one line in the format: \\nSCORE:x where x is the float score (e.g. SCORE:4.7)\\n`\\n\\nWe find that in this setup, Exa is state of the art in end to end retrieval.\\n\\n### Future Work\\n\\nWe recognize that these are only a small, incomplete set of the possible benchmarks to run. We plan on curating and releasing many more benchmarks to test search quality in challenging, real world use cases, as well as adapting more benchmarks from the rich IR literature for black-box evaluation of public search providers.\\n\\nSearch requests are extremely diverse and it’s difficult to find public datasets that capture all that diversity. Check out our Websets eval f\\n\\n[Content truncated — showing first 5,000 of 5,534 chars. LLM summarization timed out. To fix: increase auxiliary.web_extract.timeout in config.yaml, or use a faster auxiliary model. Use browser_navigate for the full page.]\",\n \"error\": null\n }\n ]\n}",
"tool_call_id": "web_extract:14"
},
{
"role": "assistant",
"content": "**Both work!**\n\n`web_extract` successfully pulled content from both URLs:\n\n1. **exa.ai/evals** - Got the full page with their benchmark posts list\n2. **exa.ai/blog/api-evals** - Got the detailed blog post about their SimpleQA and MSMARCO benchmarks (content was truncated at 5,000 chars but that's a config limit, not an Exa issue)\n\n**Summary:**\n- `web_search` ✅ - Uses Exa as backend, returns results\n- `web_extract` ✅ - Also uses Exa, extracts page content\n\nYour Exa integration is working. The config shows `web: backend: exa` which means both web tools route through Exa instead of the default (usually DDG or similar).",
"reasoning": "Both web_search and web_extract are working with Exa. The user now has confirmation that both tools are functioning properly."
},
{
"role": "user",
"content": "and how about honcho? does it work as expected?"
},
{
"role": "assistant",
"content": "Let me test the Honcho tools:",
"tool_calls": [
{
"id": "honcho_profile:15",
"call_id": "honcho_profile:15",
"response_item_id": "fc_honcho_profile15",
"type": "function",
"function": {
"name": "honcho_profile",
"arguments": "{}"
}
},
{
"id": "honcho_search:16",
"call_id": "honcho_search:16",
"response_item_id": "fc_honcho_search16",
"type": "function",
"function": {
"name": "honcho_search",
"arguments": "{\"query\": \"preferences music apps\"}"
}
}
],
"reasoning": "The user is asking about Honcho - whether it works as expected. Let me check the memory context and see what Honcho tools are available and if they're working.\n\nFrom the system, I can see:\n- `honcho_profile` - Get user's peer card\n- `honcho_search` - Semantic search over Honcho's memory\n- `honcho_context` - Ask Honcho a natural language question\n- `honcho_conclude` - Write a conclusion about the user\n\nThe memory context mentions \"The LLM cron job switches at 07:00 UTC on April 5th\" which seems like it came from Honcho.\n\nLet me test if Honcho is working by calling some of these tools."
},
{
"role": "tool",
"content": "{\"result\": \"No profile facts available yet.\"}",
"tool_call_id": "honcho_profile:15"
},
{
"role": "tool",
"content": "{\"result\": \"## Explicit Observations\\n\\n[2026-04-10 03:41:47] user-default-ubuntu has OpenRouter API access with a $0.53 backup balance.\\n[2026-04-10 03:41:47] user-default-ubuntu prefers MEMORIES.md for curated long-term memory during session startup.\\n[2026-04-10 03:41:47] user-default-ubuntu does not use Apple devices and considers Apple skills as non-priority.\\n[2026-04-10 03:41:47] user-default-ubuntu clarified that a skill to control YouTube Music is not necessary for now.\\n[2026-04-10 03:41:47] user-default-ubuntu does not want Pipo to interact with other people via WhatsApp/messaging for now.\\n[2026-04-10 03:41:47] user-default-ubuntu lives in Argentina, and scraping local sites may require an Argentinian exit node (Pixel 6a via Tailscale).\\n[2026-04-10 03:41:47] user-default-ubuntu wants audio only when explicitly requested.\\n[2026-04-10 03:41:47] user-default-ubuntu's preferred names are Igna or Chicho.\\n[2026-04-10 03:41:47] user-default-ubuntu does not rely on exec just to list files that are already known.\\n[2026-04-10 03:41:47] user-default-ubuntu has DeepSeek API access configured in Hermes, with the key in Bitwarden.\\n[2026-04-10 03:41:47] user-default-ubuntu has Kimi K2.5 API access.\\n[2026-04-10 03:41:47] user-default-ubuntu requested responses in voice and was responded to with TTS.\\n[2026-04-10 03:41:47] user-default-ubuntu's canonical long-term memory file in this workspace is MEMORIES.md.\\n[2026-04-10 03:41:47] user-default-ubuntu uses memory/YYYY-MM-DD.md files for recent context.\\n[2026-04-10 03:41:47] user-default-ubuntu has ClawRouter with wallet 0x1d1776D... and $11.70 Base balance.\\n[2026-04-10 03:41:47] user-default-ubuntu does not intend to use DeepSeek or Kimi as a daily driver, only for workers/sub-agents.\\n[2026-04-10 03:41:47] user-default-ubuntu's prior memory file exists for backward compatibility with older startup routines.\\n[2026-04-10 03:41:47] user-default-ubuntu does not use Spotify; prefers YouTube Music.\\n[2026-04-10 03:41:47] user-default-ubuntu has Mistral API access with a budget of $10.\\n[2026-04-10 03:42:31] user-default-ubuntu asked someone if 'exa' works for them and requested they try it.\\n\"}",
"tool_call_id": "honcho_search:16"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "honcho_context:17",
"call_id": "honcho_context:17",
"response_item_id": "fc_honcho_context17",
"type": "function",
"function": {
"name": "honcho_context",
"arguments": "{\"query\": \"What API keys and services does the user have configured?\"}"
}
}
],
"reasoning": "Interesting - Honcho is working! The honcho_search returned results, but honcho_profile shows no profile facts yet. Let me also try honcho_context to see if that works."
},
{
"role": "tool",
"content": "{\"result\": \"The user has the following API access configured:\\n\\n* **Kimi K2.5 API**\\n* **Mistral API** with a budget of $10.\\n* **DeepSeek API** (key is in Bitwarden)\\n* **OpenRouter API** with a $0.53 backup balance.\\n* **ClawRouter** with wallet `0x1d1776D...` and a $11.70 Base balance.\\n\\nAdditionally, their default LLM until April 4th was `claude-sonnet-4-6` via Anthropic OAuth. They have also requested the ability to change Agent Profile defaults (model, provider, skills, API keys) via WebUI.\"}",
"tool_call_id": "honcho_context:17"
},
{
"role": "assistant",
"content": "**Honcho is working!** Here's what I found:\n\n| Tool | Status | Result |\n|------|--------|--------|\n| `honcho_profile` | ⚠️ Empty | No profile facts stored yet |\n| `honcho_search` | ✅ Working | Found 18 observations about you |\n| `honcho_context` | ✅ Working | Synthesized answer about your APIs |\n\nThe search found facts like:\n- You prefer YouTube Music (no Spotify)\n- Lives in Argentina, scraping may need Pixel 6a exit node\n- DeepSeek key in Bitwarden, Kimi K2.5, Mistral ($10), OpenRouter ($0.53 backup)\n- ClawRouter wallet with $11.70 Base\n- Don't want me interacting with others via WhatsApp\n- Preferred names: Igna or Chicho\n\nLet me test writing to Honcho too:",
"tool_calls": [
{
"id": "honcho_conclude:18",
"call_id": "honcho_conclude:18",
"response_item_id": "fc_honcho_conclude18",
"type": "function",
"function": {
"name": "honcho_conclude",
"arguments": "{\"conclusion\": \"User confirmed Exa web search and extract tools are working as expected on April 10, 2026.\"}"
}
}
],
"reasoning": "Great! Honcho is working. Let me test honcho_conclude to make sure write operations work too."
},
{
"role": "tool",
"content": "{\"result\": \"Conclusion saved: User confirmed Exa web search and extract tools are working as expected on April 10, 2026.\"}",
"tool_call_id": "honcho_conclude:18"
},
{
"role": "assistant",
"content": "**`honcho_conclude` also works** - just saved that conclusion.\n\n**Summary: Honcho is fully functional**\n- ✅ `honcho_search` - retrieves raw observations\n- ✅ `honcho_context` - synthesizes answers from stored facts\n- ✅ `honcho_conclude` - writes new conclusions\n- ⚠️ `honcho_profile` - works but empty (no curated profile facts yet)\n\nThe hybrid mode is working - facts get auto-injected into my context (that's how I knew about the cron job switching at 07:00 UTC on April 5th).",
"reasoning": "Honcho conclude works too. All Honcho tools are functional."
},
{
"role": "user",
"content": "alright I have now enabled you parallel, openai for tts as well as elevenlabs and github, care to confirm?"
},
{
"role": "assistant",
"content": "Let me check what's now configured and test each one:",
"reasoning": "The user says they've enabled several new integrations:\n1. Parallel - probably Parallel web search/extract\n2. OpenAI for TTS\n3. ElevenLabs\n4. GitHub\n\nI should check the config to see what's enabled and test these tools to confirm they work.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "read_file:19",
"call_id": "read_file:19",
"response_item_id": "fc_read_file19",
"type": "function",
"function": {
"name": "read_file",
"arguments": "{\"path\": \"~/.hermes/config.yaml\", \"offset\": 200, \"limit\": 50}"
}
},
{
"id": "terminal:20",
"call_id": "terminal:20",
"response_item_id": "fc_terminal20",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"grep -A2 \\\"tts:\\\" ~/.hermes/config.yaml | head -20\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"content\": \" 200| redact_pii: false\\n 201|tts:\\n 202| provider: edge\\n 203| edge:\\n 204| voice: en-US-AriaNeural\\n 205| elevenlabs:\\n 206| voice_id: pNInz6obpgDQGcFmaJgB\\n 207| model_id: eleven_multilingual_v2\\n 208| openai:\\n 209| model: gpt-4o-mini-tts\\n 210| voice: alloy\\n 211| neutts:\\n 212| ref_audio: ''\\n 213| ref_text: ''\\n 214| model: neuphonic/neutts-air-q4-gguf\\n 215| device: cpu\\n 216|stt:\\n 217| enabled: true\\n 218| provider: local\\n 219| local:\\n 220| model: base\\n 221| language: ''\\n 222| openai:\\n 223| model: whisper-1\\n 224| mistral:\\n 225| model: voxtral-mini-latest\\n 226| model: whisper-1\\n 227|voice:\\n 228| record_key: ctrl+b\\n 229| max_recording_seconds: 120\\n 230| auto_tts: false\\n 231| silence_threshold: 200\\n 232| silence_duration: 3\\n 233|human_delay:\\n 234| mode: false\\n 235| min_ms: 800\\n 236| max_ms: 2500\\n 237|memory:\\n 238| memory_enabled: true\\n 239| user_profile_enabled: true\\n 240| memory_char_limit: 2200\\n 241| user_char_limit: 1375\\n 242| provider: honcho\\n 243| flush_min_turns: 6\\n 244| nudge_interval: 10\\n 245|delegation:\\n 246| model: ''\\n 247| provider: ''\\n 248| base_url: ''\\n 249| api_key: ''\\n 250|\", \"total_lines\": 437, \"file_size\": 11529, \"truncated\": true, \"hint\": \"Use offset=250 to continue reading (showing 200-249 of 437 lines)\", \"is_binary\": false, \"is_image\": false}",
"tool_call_id": "read_file:19"
},
{
"role": "tool",
"content": "{\"output\": \"tts:\\n provider: edge\\n edge:\\n--\\n neutts:\\n ref_audio: ''\\n ref_text: ''\\n--\\n auto_tts: false\\n silence_threshold: 200\\n silence_duration: 3\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:20"
},
{
"role": "assistant",
"content": "",
"reasoning": "I see TTS config is there. I need to check:\n1. Web backend for Parallel (currently set to exa)\n2. GitHub integration\n3. TTS providers\n\nLet me check the web config and search for parallel/github.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "terminal:21",
"call_id": "terminal:21",
"response_item_id": "fc_terminal21",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"grep -i \\\"parallel\\\\|github\\\" ~/.hermes/config.yaml\"}"
}
},
{
"id": "execute_code:22",
"call_id": "execute_code:22",
"response_item_id": "fc_execute_code22",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\n# Check what TTS providers are available and if their API keys are set\\nimport os\\n\\nproviders = {\\n 'OPENAI_API_KEY': 'OpenAI',\\n 'ELEVENLABS_API_KEY': 'ElevenLabs',\\n}\\n\\nprint(\\\"TTS Provider API Keys:\\\")\\nfor key, name in providers.items():\\n val = os.environ.get(key)\\n status = f\\\"✅ Present ({val[:8]}...)\\\" if val else \\\"❌ Not set\\\"\\n print(f\\\" {name}: {status}\\\")\\n\\n# Check web backend for Parallel\\nprint(f\\\"\\\\nWeb backend: exa (Parallel not directly visible here)\\\")\\n\\n# Check for GitHub\\ngh_token = os.environ.get('GITHUB_TOKEN') or os.environ.get('GH_TOKEN')\\nprint(f\\\"GitHub token: {'✅ Present' if gh_token else '❌ Not set'}\\\")\\n\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"output\": \"\", \"exit_code\": 1, \"error\": null}",
"tool_call_id": "terminal:21"
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"TTS Provider API Keys:\\n OpenAI: ❌ Not set\\n ElevenLabs: ❌ Not set\\n\\nWeb backend: exa (Parallel not directly visible here)\\nGitHub token: ❌ Not set\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.21}",
"tool_call_id": "execute_code:22"
},
{
"role": "assistant",
"content": "",
"reasoning": "Hmm, the API keys aren't in the environment. Let me check the auxiliary section where web_extract backend might be configured, and also check if there are mcp_servers or other configurations for GitHub and Parallel.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "read_file:23",
"call_id": "read_file:23",
"response_item_id": "fc_read_file23",
"type": "function",
"function": {
"name": "read_file",
"arguments": "{\"path\": \"~/.hermes/config.yaml\", \"offset\": 130, \"limit\": 50}"
}
},
{
"id": "terminal:24",
"call_id": "terminal:24",
"response_item_id": "fc_terminal24",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"grep -A5 \\\"mcp_servers:\\\" ~/.hermes/config.yaml\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"content\": \" 130| cheap_model:\\n 131| model: qwen/qwen3.6-plus:free\\n 132| provider: openrouter\\n 133|auxiliary:\\n 134| vision:\\n 135| provider: auto\\n 136| model: ''\\n 137| base_url: ''\\n 138| api_key: ''\\n 139| timeout: 30\\n 140| download_timeout: 30\\n 141| web_extract:\\n 142| provider: auto\\n 143| model: ''\\n 144| base_url: ''\\n 145| api_key: ''\\n 146| timeout: 30\\n 147| compression:\\n 148| provider: auto\\n 149| model: ''\\n 150| base_url: ''\\n 151| api_key: ''\\n 152| timeout: 120\\n 153| session_search:\\n 154| provider: auto\\n 155| model: ''\\n 156| base_url: ''\\n 157| api_key: ''\\n 158| timeout: 30\\n 159| skills_hub:\\n 160| provider: auto\\n 161| model: ''\\n 162| base_url: ''\\n 163| api_key: ''\\n 164| timeout: 30\\n 165| approval:\\n 166| provider: auto\\n 167| model: ''\\n 168| base_url: ''\\n 169| api_key: ''\\n 170| timeout: 30\\n 171| mcp:\\n 172| provider: auto\\n 173| model: ''\\n 174| base_url: ''\\n 175| api_key: ''\\n 176| timeout: 30\\n 177| flush_memories:\\n 178| provider: auto\\n 179| model: ''\\n 180|\", \"total_lines\": 437, \"file_size\": 11529, \"truncated\": true, \"hint\": \"Use offset=180 to continue reading (showing 130-179 of 437 lines)\", \"is_binary\": false, \"is_image\": false}",
"tool_call_id": "read_file:23"
},
{
"role": "tool",
"content": "{\"output\": \"mcp_servers:\\n context7:\\n command: npx\\n args:\\n - -y\\n - '@upstash/context7-mcp@latest'\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:24"
},
{
"role": "assistant",
"content": "Let me test the tools directly to see if they work:",
"reasoning": "I don't see Parallel, GitHub, or the TTS API keys configured yet. The user said they \"enabled\" them but maybe they mean they just added the configs or the API keys aren't loaded in my environment yet. Let me check the full config for any new changes and also check if there's a web backend setting or separate parallel configuration.\n\nActually, I should reload the config or check if there's been any recent changes. Maybe the user updated the config but I need to verify what's actually there now.\n\nLet me also try to actually USE the tools to see if they work - that would be the real test.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "text_to_speech:25",
"call_id": "text_to_speech:25",
"response_item_id": "fc_text_to_speech25",
"type": "function",
"function": {
"name": "text_to_speech",
"arguments": "{\"text\": \"Testing TTS\", \"output_path\": \"/tmp/test_tts.mp3\"}"
}
},
{
"id": "web_search:26",
"call_id": "web_search:26",
"response_item_id": "fc_web_search26",
"type": "function",
"function": {
"name": "web_search",
"arguments": "{\"query\": \"test parallel\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"success\": true, \"file_path\": \"/tmp/test_tts.ogg\", \"media_tag\": \"[[audio_as_voice]]\\nMEDIA:/tmp/test_tts.ogg\", \"provider\": \"edge\", \"voice_compatible\": true}",
"tool_call_id": "text_to_speech:25"
},
{
"role": "tool",
"content": "{\n \"success\": true,\n \"data\": {\n \"web\": [\n {\n \"url\": \"https://www.qawolf.com/blog/what-is-parallel-software-testing\",\n \"title\": \"What is Parallel Software Testing? The Complete Guide to Faster Test Execution | QA Wolf\",\n \"description\": \"What is Parallel Software Testing? The Complete Guide to Faster Test Execution | QA Wolf\",\n \"position\": 1\n },\n {\n \"url\": \"https://www.tricentis.com/learn/parallel-testing\",\n \"title\": \"Parallel testing: what it is and how it works - Tricentis\",\n \"description\": \"Parallel testing: what it is and how it works - Tricentis\\n\\nSkip to content\\n\\nCompany Back Company Management team Careers News Locations Partners Blog Customer Portal\\n\\nHome\\n\\nLearn\\n\\nDevOps Parallel testing: What it is and how it works\\n\\n## Learn\\n\\n# Parallel testing: what it is and how it works Learn what parallel testing is, how it works, and why QA teams use it to improve coverage, efficiency, and pipeline performance. One of the biggest challenges with software testing is the time that it takes. As Stephan Schmidt ... One way to tackle this problem is by adopting parallel testing. In this post, we’re going to break down what parallel testing is, why it’s valuable for your team, how you can adopt it, and what mistakes you need to avoid to get the most out of it. ## What is parallel testing? On the surface, parallel testing is a pretty simple concept: run your tests in parallel! Or, to expand a little, utilize multiple CPU cores to execute tests at the same time as one another. Let’s consider an application with two test files: one tests a validator function for email addresses and the other tests some calendar-based scheduling logic. The email tests take ten seconds to run, and the calendar tests take 45 seconds. When run in sequence, those tests will take 55 seconds to run. But, with even a little bit of clever work, we can ... Executed perfectly, the parallel workflow would still be just 46 seconds! By running independent tests in parallel, we ... In real-world scenarios, no team is optimizing for 10 seconds of quicker test execution. But most companies that invest ... like 55 minutes. On the surface, parallel testing is a pretty simple concept: run your tests in parallel! ... As we already noted, the primary benefit of adopting parallel tests is the execution performance. If you run your tests ... ## How does parallel testing work? We’ve already covered the simplest form of parallel testing. You have a small setup script that forks a number of testing processes and then waits for them all to finish. If they all succeed, you report that back to the user. If not, you report that back to the user, too. In reality, you probably don’t need to do this yourself. It’s likely that your testing library already comes with the ... If you’re running tests across multiple languages, it might be worth looking into a dedicated parallel testing tool. A library like btest provides the ability to configure your parallel tests across all your languages, meaning ... Those tools are good fits for small and medium-sized codebases. If you’re working on a bigger codebase, you might want ... Tricentis Testim test automation not only makes authoring new tests faster and easier, it also makes ... ## What are some key considerations with parallel testing? If you’re considering adopting parallel tests, there are some things you should keep in mind: ... ## Tricentis Tosca is here for your testing needs By now, the benefits of parallel testing should be obvious. If you’ve committed to automated testing as part of your ... You might already be there! If you are, Tosca is right here to help. Our [continuous test automation ... This post was written by Eric Boersma. Eric is a software\",\n \"position\": 2\n },\n {\n \"url\": \"https://playwright.dev/docs/test-parallel\",\n \"title\": \"Parallelism | Playwright\",\n \"description\": \"Parallelism | Playwright\\n\\nOn this page\\n\\n## Introduction Playwright Test runs tests in parallel. In order to achieve that, it runs several worker processes that run at the same time. By default, test files are run in parallel. Tests in a single file are run in order, in the same worker process. - You can configure tests using`test.describe.configure` to run tests in a single file in parallel. - You can configure entire project to have all tests in all files to run in parallel using testProject.fullyParallel or testConfig.fullyParallel. - To disable parallelism limit the number of workers to one. You can control the number of parallel worker processes and limit the number of failures in the whole test suite for ... All tests run in worker processes. These processes are OS processes, running independently, orchestrated by the test ... You can't communicate between the workers. Playwright Test reuses a single worker as much as it can to make testing ... ## Limit workers You can control the maximum number of parallel worker processes via command line or in the [configuration ... From the command line:\\n\\n```bash\\nnpx playwright test --workers 4\\n```\\n\\nIn the configuration file:\\n\\nplaywright.config.ts ... ## Disable parallelism You can disable any parallelism by allowing just a single worker at any time. Either set`workers: 1` option in the configuration file or pass`--workers=1` to the command line. ```bash\\nnpx playwright test --workers=1\\n```\\n\\n## Parallelize tests in a single file By default, tests in a single file are run in order. If you have many independent tests in a single file, you might want to run them in parallel with test.describe.configure(). Note that parallel tests are executed in separate worker processes and cannot share any state or global variables. Each test executes all relevant hooks just for itself, including`beforeAll` and`afterAll`. ```js import { test } from '@playwright/test';test.describe.configure({ mode: 'parallel' });test('runs in parallel 1', async ... Alternatively, you can opt-in all tests into this fully-parallel mode in the configuration file:\\n\\nplaywright.config.ts ```js\\nimport { defineConfig } from '@playwright/test';export default defineConfig({ fullyParallel: true,});\\n``` You can also opt in for fully-parallel mode for just a few projects:\\n\\nplaywright.config.ts ... import { defineConfig } from '@playwright/test';export default defineConfig({ // runs all tests in all files of a specific project in parallel projects: [ { name: 'chromium', use: { ...devices['Desktop Chrome'] }, ... ## Serial mode You can annotate inter-dependent tests as serial. If one of the serial tests fails, all subsequent tests are skipped. ... ## Worker index and parallel index Each worker process is assigned two ids: a unique worker index that starts with 1, and a parallel index that is between`0` and`workers - 1`. When a worker is restarted, for example after a failure, the new worker process has the ... ## Control test order Playwright Test runs tests from a single file in the order of declaration, unless you parallelize tests in a single ... There is no guarantee about the order of test execution across the files, because Playwright Test runs test files in parallel by default. However, if you disable parallelism, you can control test order by either naming your files in ... Now disable parallel execution by setting workers to one, and specify your test list file.\\n\\nplaywright.config.ts ... Worker index and parallel index\\n\\n- Isolate\",\n \"position\": 3\n },\n {\n \"url\": \"https://vitest.dev/guide/parallelism\",\n \"title\": \"Parallelism | Guide | Vitest\",\n \"description\": \"Parallelism | Guide | Vitest\\n\\nAnnouncing Vite+ Alpha: Open source. Unified. Next-gen.\\n\\n# Parallelism ## File Parallelism By default, Vitest runs test files in parallel. Depending on the specified`pool`, Vitest uses a different mechanism to parallelize test files: - `forks`(the default) and`vmForks` run tests in different child processes - `threads` and`vmThreads` run tests in different worker threads Both \\\"child processes\\\" and \\\"worker threads\\\" are referred to as \\\"workers\\\". You can configure the number of running workers with maxWorkers option. If you have a lot of tests, it is usually faster to run them in parallel, but it also depends on the project, the environment and isolation state. To disable file parallelisation, you can set fileParallelism to`false`. To learn more about possible performance improvements, read the Performance Guide. ## Test Parallelism Unlike test files, Vitest runs tests in sequence. This means that tests inside a single test file will run in the order they are defined. Vitest supports the concurrent option to run tests together. If this option is set, Vitest will group concurrent tests in the same file (the number of simultaneously running tests depends on the maxConcurrency option) and run them with Promise.all. The hook execution order within a single group is also controlled by sequence.hooks. With`sequence.hooks: 'parallel'`, the execution is bounded by the same limit of maxConcurrency. Vitest doesn't perform any smart analysis and doesn't create additional workers to run these tests. This means that the performance of your tests will improve only if you rely heavily on asynchronous operations. For example, these tests will still run one after another even though the`concurrent` option is specified. This is because they are synchronous: ts\\n\\n```\\ntest.concurrent('the first test', () => {\\n expect(1).toBe(1)\\n}) test.concurrent('the second test', () => {\\n expect(2).toBe(2)\\n})\\n``` If you wish to run all tests concurrently, you can set the [sequence.concurrent\",\n \"position\": 4\n },\n {\n \"url\": \"https://circleci.com/docs/guides/optimize/parallelism-faster-jobs/index.md\",\n \"title\": \"Test splitting and parallelism\",\n \"description\": \"# Test splitting and parallelism\\n\\nUse parallelism and test splitting to: - Reduce the time taken for the testing portion of your CI/CD pipeline. - Specify a number of executors across which to split your tests. - Split your test suite using one of the options provided by the CircleCI CLI: by name, size or by using timing data. If you are interested to read about concurrent job runs, see the Concurrency overview page. ... Pipelines are often configured so that a set of tests are run each time code is committed. The more tests your project has, the longer it will take for them to complete using a single compute resource. To reduce this time, you can split your tests and run them across multiple, parallel-running execution environments. Test splitting is a great way to speed up the testing portion of your CI/CD pipeline. CircleCI test splitting lets you intelligently define where splits happen across a test suite: ... It is also possible to use the CLI to manually allocate tests across parallel environments.\\n\\n## Quickstart The following how-to guide walks you through the following: - Steps required to glob your test files.\\n- Split your tests evenly across parallel containers or VMs. - Run your tests in parallel. How-to: Use the CircleCI CLI to split tests\\n\\n## Specify parallel environments for a test job Test suites are conventionally defined at the job level in your `.circleci/config.yml` file. The `parallelism` key specifies how many independent executors are set up to run the job. To run a job’s steps in multiple, parallel execution environments, set the `parallelism` key to a value greater than `1`. In the example below, `parallelism` is set to `4`, meaning four identical execution environments will be set up ... Use the `circleci tests run` command to split and run your tests. Your tests will be split up, and a portion of your ... ```\\n# ~/.circleci/config.yml\\njobs:\\n build:\\n docker:\\n - image: cimg/go:1.18.1\\n parallelism: 4 ... To use parallelism with jobs that use self-hosted runners, ensure that you have at least two runners associated with the resource class that your job uses. If you set the parallelism value to be greater than the number ... For more information, see the Configuration reference page.\\n\\n## How test splitting works CircleCI’s intelligent test splitting allows you to: - Specify a number of identical execution environments for a job. - Split your test suite into buckets so that the tests can be run in parallel on your configured execution environments. Test splitting reduces the length of time taken for testing in your pipelines. You can split your tests up in a few different ways: - Alphabetically, by name. This is the default:# ~/.circleci/config.yml\\njobs:\\n build:\\n docker: - image: cimg/go:1.18.1\\n parallelism: 4\\n resource_class: large\\n steps: ... Using timing-based test splitting, timing data from the previous test run is used to split a test suite as evenly as ... 1. Introduce parallelism to spin up a number of identical test environments (4 in this example) ... Use the `circleci tests run` command to run your tests, split your tests across parallel executors and take advantage ... ```\\n\\n## Manual allocation For full control over how tests are split across parallel executors, CircleCI provides two environment variables that you can use in place of the CLI to configure each container individually. - `$CIRCLE_NODE_TOTAL` is the total number of parallel containers being used to run your job. ... The CLI looks up the number of available execution environments (`$CIRCLE_NODE_TOTAL`), along with the current container index (`$CIRCLE_NODE_INDEX`). Then, it uses deterministic splitting algorithms to split the test files across ... The number\",\n \"position\": 5\n }\n ]\n }\n}",
"tool_call_id": "web_search:26"
},
{
"role": "assistant",
"content": "TTS works (used Edge provider). Let me check GitHub and test if Parallel is configured:",
"reasoning": "TTS works (used edge provider by default). Web search still returns results through Exa (not Parallel specifically).\n\nNow let me check if GitHub is configured by trying to use it, and also look for Parallel configuration more specifically.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "terminal:27",
"call_id": "terminal:27",
"response_item_id": "fc_terminal27",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"gh auth status 2>&1 || echo \\\"GitHub CLI not authenticated\\\"\"}"
}
},
{
"id": "execute_code:28",
"call_id": "execute_code:28",
"response_item_id": "fc_execute_code28",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\n# Check if there are mcp servers for github or parallel\\nimport subprocess\\nresult = subprocess.run(['cat', '/home/ubuntu/.hermes/config.yaml'], capture_output=True, text=True)\\nconfig = result.stdout\\n\\n# Look for parallel or github mentions\\nfor line in config.split('\\\\n'):\\n if 'parallel' in line.lower() or 'github' in line.lower():\\n print(line)\\n\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"output\": \"github.com\\n ✓ Logged in to github.com account ignaciolagosruiz (/home/ubuntu/.config/gh/hosts.yml)\\n - Active account: true\\n - Git operations protocol: https\\n - Token: gho_************************************\\n - Token scopes: 'gist', 'read:org', 'repo', 'workflow'\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:27"
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"\", \"tool_calls_made\": 0, \"duration_seconds\": 0.22}",
"tool_call_id": "execute_code:28"
},
{
"role": "assistant",
"content": "GitHub CLI is authenticated! Let me check if Parallel is set up as an MCP server or web backend:",
"reasoning": "GitHub CLI is authenticated! Let me now check if Parallel is actually configured as a web backend or as an MCP server.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "terminal:29",
"call_id": "terminal:29",
"response_item_id": "fc_terminal29",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"grep -i \\\"web:\\\" ~/.hermes/config.yaml -A2\"}"
}
},
{
"id": "execute_code:30",
"call_id": "execute_code:30",
"response_item_id": "fc_execute_code30",
"type": "function",
"function": {
"name": "execute_code",
"arguments": "{\"code\": \"\\n# Check the entire config for parallel-related entries\\nimport re\\n\\nwith open('/home/ubuntu/.hermes/config.yaml') as f:\\n content = f.read()\\n\\n# Find all lines mentioning parallel, exa, or search backend\\nlines = content.split('\\\\n')\\nfor i, line in enumerate(lines, 1):\\n if any(x in line.lower() for x in ['parallel', 'exa', 'backend', 'search']):\\n print(f\\\"{i}: {line}\\\")\\n\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"output\": \"web:\\n backend: exa\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:29"
},
{
"role": "tool",
"content": "{\"status\": \"success\", \"output\": \"64: deeper meaning behind every query. Let us examine not just the 'how' but the\\n77: teacher: You are a patient teacher. Explain concepts clearly with examples.\\n87: backend: local\\n153: session_search:\\n342: - session_search\\n360: - session_search\\n383: - session_search\\n404: backend: exa\\n\", \"tool_calls_made\": 0, \"duration_seconds\": 0.22}",
"tool_call_id": "execute_code:30"
},
{
"role": "assistant",
"content": "",
"reasoning": "I see `web: backend: exa` but no Parallel configuration. The user said they enabled Parallel but maybe they haven't added it to the config yet, or they mean something else.\n\nLet me check if there are any MCP servers beyond context7 that might include Parallel or GitHub.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "read_file:31",
"call_id": "read_file:31",
"response_item_id": "fc_read_file31",
"type": "function",
"function": {
"name": "read_file",
"arguments": "{\"path\": \"~/.hermes/config.yaml\", \"offset\": 318, \"limit\": 30}"
}
}
]
},
{
"role": "tool",
"content": "{\"content\": \" 318| timeout: 300\\n 319|group_sessions_per_user: true\\n 320|mcp_servers:\\n 321| context7:\\n 322| command: npx\\n 323| args:\\n 324| - -y\\n 325| - '@upstash/context7-mcp@latest'\\n 326| env:\\n 327| CONTEXT7_API_KEY: ctx7sk-457a3328-3870-45bb-849e-808a8b9b6666\\n 328| enabled: true\\n 329|platform_toolsets:\\n 330| api_server:\\n 331| - hermes-api-server\\n 332| cli:\\n 333| - browser\\n 334| - clarify\\n 335| - code_execution\\n 336| - cronjob\\n 337| - delegation\\n 338| - file\\n 339| - image_gen\\n 340| - memory\\n 341| - moa\\n 342| - session_search\\n 343| - skills\\n 344| - terminal\\n 345| - todo\\n 346| - tts\\n 347| - vision\\n 348|\", \"total_lines\": 437, \"file_size\": 11529, \"truncated\": true, \"hint\": \"Use offset=348 to continue reading (showing 318-347 of 437 lines)\", \"is_binary\": false, \"is_image\": false}",
"tool_call_id": "read_file:31"
},
{
"role": "assistant",
"content": "",
"reasoning": "I only see `context7` in mcp_servers. Let me check the rest of the file and also check if there's a parallel mcp server that should be added.",
"finish_reason": "tool_calls",
"tool_calls": [
{
"id": "terminal:32",
"call_id": "terminal:32",
"response_item_id": "fc_terminal32",
"type": "function",
"function": {
"name": "terminal",
"arguments": "{\"command\": \"wc -l ~/.hermes/config.yaml && tail -50 ~/.hermes/config.yaml\"}"
}
}
]
},
{
"role": "tool",
"content": "{\"output\": \"437 /home/ubuntu/.hermes/config.yaml\\n - vision\\n - web\\n whatsapp:\\n - hermes-whatsapp\\nplatforms:\\n api_server:\\n enabled: true\\n host: 0.0.0.0\\n port: 8642\\nsession_reset:\\n at_hour: 4\\n idle_minutes: 1440\\n mode: none\\nstreaming:\\n enabled: false\\nweb:\\n backend: exa\\n\\n# ── Fallback Model ────────────────────────────────────────────────────\\n# Automatic provider failover when primary is unavailable.\\n# Uncomment and configure to enable. Triggers on rate limits (429),\\n# overload (529), service errors (503), or connection failures.\\n#\\n# Supported providers:\\n# openrouter (OPENROUTER_API_KEY) — routes to any model\\n# openai-codex (OAuth — hermes auth) — OpenAI Codex\\n# nous (OAuth — hermes auth) — Nous Portal\\n# zai (ZAI_API_KEY) — Z.AI / GLM\\n# kimi-coding (KIMI_API_KEY) — Kimi / Moonshot\\n# minimax (MINIMAX_API_KEY) — MiniMax\\n# minimax-cn (MINIMAX_CN_API_KEY) — MiniMax (China)\\n#\\n# For custom OpenAI-compatible endpoints, add base_url and api_key_env.\\n#\\n# fallback_model:\\n# provider: openrouter\\n# model: anthropic/claude-sonnet-4\\n#\\n# ── Smart Model Routing ────────────────────────────────────────────────\\n# Optional cheap-vs-strong routing for simple turns.\\n# Keeps the primary model for complex work, but can route short/simple\\n# messages to a cheaper model across providers.\\n#\\n# smart_model_routing:\\n# enabled: true\\n# max_simple_chars: 160\\n# max_simple_words: 28\\n# cheap_model:\\n# provider: openrouter\\n# model: google/gemini-2.5-flash\", \"exit_code\": 0, \"error\": null}",
"tool_call_id": "terminal:32"
},
{
"role": "assistant",
"content": "Here's what I found:\n\n| Integration | Status | Details |\n|-------------|--------|---------|\n| **GitHub** | ✅ Working | `gh` CLI authenticated as ignaciolagosruiz with repo/gist/workflow scopes |\n| **TTS - Edge** | ✅ Working | Default provider, used successfully |\n| **TTS - OpenAI** | ⚠️ Configured but no API key | Config present (gpt-4o-mini-tts, voice: alloy) but `OPENAI_API_KEY` not in env |\n| **TTS - ElevenLabs** | ⚠️ Configured but no API key | Config present (multilingual_v2) but `ELEVENLABS_API_KEY` not in env |\n| **Parallel** | ❓ Not visible | Web backend still set to `exa`, no Parallel MCP server found |\n\n**For Parallel** - Is it meant to be:\n- A web backend replacement for Exa?\n- An MCP server?\n- Something else?\n\n**For TTS API keys** - They might need to be loaded into my environment or set in the config under the `tts.openai.api_key` / `tts.elevenlabs.api_key` fields.\n\nWant me to add/fix any of these?",
"reasoning": "Okay, so I see the full config now. The user said they enabled Parallel, OpenAI TTS, ElevenLabs, and GitHub. Let me summarize what I found:\n\n1. **Parallel** - Not visible in the config yet. Web backend is still set to `exa`\n2. **OpenAI TTS** - Config is present (lines 208-210) but API key not in env\n3. **ElevenLabs TTS** - Config is present (lines 205-207) but API key not in env \n4. **GitHub** - GitHub CLI is authenticated ✅\n\nMaybe the user means they just added the TTS configs and GitHub auth, but Parallel might need to be added differently. Or maybe \"Parallel\" is the company/product name and I should check if it's set up as an MCP server.\n\nLet me verify the current state clearly for the user.",
"finish_reason": "stop"
}
]
}