What defines the framework
Receives your message, thinks through what's needed, uses the right tools, and repeats until the task is complete — all without step-by-step guidance.
Remembers your conversations, preferences, and important facts across sessions. Everything is stored as human-readable files you can inspect and edit.
The agent can write its own skill files to learn new abilities. Skills are simple instruction documents that shape how the agent approaches specific tasks.
Runs on a schedule (every 30 minutes by default) to check a task list and take action when needed — without being prompted. Create recurring jobs anytime.
Talk to your agent via Telegram, Discord, web chat, or command line. All channels use the same interface — adding new platforms is straightforward.
A built-in web interface to control everything: toggle security settings, manage tool permissions, configure integrations, and view audit logs.
Choose how you want to communicate
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.
Swap AI providers anytime
Full tool-use support with the most capable reasoning. The recommended choice for complex autonomous tasks.
Compatible via the OpenAI API adapter. Use any GPT model you have access to.
Run completely offline with local models. No API costs, full privacy, works without internet.
FREE & OFFLINE31 tools across 7 categories
All protections 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.
Guided setup in under 5 minutes
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.
Technical parameters