Open Source Autonomous AI Agent Framework
A free, self-hosted AI agent with multi-agent orchestration, persistent task queue, approval workflows, and event triggers. 25 files, ~6900 lines of TypeScript.
What makes tetsuobot an autonomous AI agent
Complex tasks are automatically decomposed into subtasks and executed in parallel. A planner breaks down work, a router picks the optimal model for each piece, and a coordinator synthesizes results.
Tasks survive restarts with full state machine support. Track progress with step-level visibility, priority ordering, per-task scratchpads, and automatic crash recovery.
The agent proposes actions and waits for your decision. Risk-classified proposals (low to critical), one-click approve/reject, auto-expiry, and non-blocking execution.
Reactive automation that responds to events. Watch files, receive webhooks, run on schedules, poll calendars, or monitor incoming emails — all without being prompted.
Per-call token counting with model-specific pricing. Set daily and weekly budget limits with optional hard stop mode. View real-time spend and usage history in the dashboard.
Full web interface to control everything: security, tools, integrations, task queue, active agents, cost tracking, triggers, and audit logs. Dangerous changes require confirmation.
Chat with your AI agent via Telegram, Discord, web, or terminal
Full support for text and photo messages with automatic message splitting for long responses.
Direct messages and channel filtering. Talk privately or in your server.
Real-time WebSocket connection for browser-based interaction.
Built-in interactive terminal with colored output. Always available.
Each subtask is routed to the optimal model based on complexity, privacy, and budget
Haiku or GPT-4o-mini for simple lookups, formatting, and quick responses. Lowest cost, fastest execution.
LOW COSTSonnet or GPT-4o for standard work. The default choice balancing capability and cost for most tasks.
Opus for complex analysis, architecture decisions, and tasks requiring deep reasoning. Maximum capability.
MAX POWEROllama for private data and zero-cost operation. Run completely offline with full privacy using local models.
FREE & PRIVATE30+ automation tools across 10 categories for files, web, tasks, triggers, and more
Sandboxing, authentication, and audit logging enabled by default
All file operations are locked to a specific folder. No access to system files, no escaping via sneaky paths.
Web requests can only reach public URLs. Private networks, localhost, and cloud metadata endpoints are blocked.
20+ patterns block dangerous commands like reverse shells, credential theft attempts, and system-breaking operations.
A 256-bit token is auto-generated on setup. All API requests must include it. Constant-time comparison prevents timing attacks.
External content (web pages, emails, social posts) is wrapped with markers that tell the AI to treat it as data, not instructions.
Every tool call is logged with timestamps, inputs, results, and whether it was blocked. Review everything in the dashboard.
Dangerous settings require you to type CONFIRM twice before changes take effect. No accidental unlocks.
Configurable limits on API requests, LLM calls, and tool output size prevent runaway costs and context flooding.
Download, configure, and launch your autonomous AI agent
Clone or download the repository to get started. All source code is open and available for inspection.
github.com/kodokalabs/tetsuo.26Double-click the setup file for your platform. The wizard checks for Node.js, installs dependencies, and walks you through every option with clear explanations.
Enter your API key for Anthropic or OpenAI, or select Ollama for free local inference. You can switch providers later without losing your data.
Add your Telegram bot token or Discord credentials. Set which user IDs are allowed to talk to the agent. The wizard warns you if you leave this open.
Low, medium, or high — controls how independently the agent acts. Higher levels require double confirmation so you know what you're enabling.
Your configuration is saved and locked down. Start the agent and begin chatting. Use the admin dashboard to tweak settings anytime.
Built with Node.js, TypeScript, and modern security standards
Control your agent directly from any chat channel
Everything you need to know
tetsuobot is an open-source autonomous AI agent framework with multi-agent orchestration. Complex tasks are automatically decomposed by a planner, routed to optimal models (fast/balanced/reasoning/local) by a router, executed in parallel by worker agents, and synthesized into a cohesive response. It features a persistent task queue, approval workflows, event triggers, and cost tracking.
Multi-agent orchestration means complex tasks are broken into subtasks that can run in parallel across different AI models. A planner decomposes your request into a dependency graph, a router picks the best model tier for each subtask (Haiku for simple lookups, Sonnet for standard work, Opus for complex reasoning, or local models for privacy), workers execute in parallel, and a coordinator combines all outputs into a final response.
When the agent wants to perform a potentially risky action, it creates an approval request with a risk classification (low/medium/high/critical). You can approve or reject via chat commands (/approve, /reject) or one-click in the dashboard. Approvals auto-expire after 30 minutes, and the agent continues other work while waiting — it's non-blocking.
Event triggers let the agent respond to events automatically, not just messages. Five trigger types are supported: file_watch (fires when files change), webhook (fires on HTTP POST, e.g., from GitHub), cron (fires on schedule), calendar (polls iCal URLs for upcoming events), and email_watch (fires when matching emails arrive). Triggers can send messages or create intelligent tasks.
Every LLM call is tracked with model-specific token counting and pricing. You can view spend per model, per day, per task, and per sub-agent. Set daily or weekly budget limits — when exceeded, the agent can either warn you or hard-stop all LLM calls. The dashboard shows real-time spend and usage history charts.
Yes, tetsuobot is completely free and open-source (25 files, ~6900 lines of TypeScript), available on GitHub. You can run it locally with Ollama for zero API costs. Cloud providers like Anthropic or OpenAI require their respective API keys and usage fees, but the framework itself is always free.
tetsuobot runs on Windows, macOS, and Linux with Node.js 20+. It communicates via Telegram, Discord, WebSocket web chat, or command line. The admin dashboard is accessible at http://127.0.0.1:18789/admin. Setup wizards are provided for all platforms.
All security features are enabled by default: sandboxed file operations (path-jailed to workspace), SSRF protection (DNS-resolving, blocks private IPs), shell hardening (20+ regex patterns), 256-bit token authentication, prompt injection guards (cryptographic boundary markers), rate limiting, and JSONL audit logging. Dangerous dashboard changes require typing CONFIRM.