Dark Mode

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

s-nagaev/chibi

Repository files navigation

Your Digital Companion. Not a Tool. A Partner.
Self-hosted, asynchronous Telegram bot that orchestrates multiple AI providers, tools, and sub-agents to get real work done.

Read this in other languages:
Espanol * Portugues (Brasil) * Ukrayins'ka * Bahasa Indonesia * Turkce * Russkii * Ri Ben Yu * Fan Ti Zhong Wen * Jian Ti Zhong Wen


Chibi is built for the moment you realize you need more than "an AI tool." You need a partner that can coordinate models, run work in the background, and integrate with your systems - without you babysitting prompts.

Chibi is an asynchronous, self-hosted Telegram-based digital companion that orchestrates multiple AI providers and tools to deliver outcomes: code changes, research syntheses, media generation, and operational tasks.


Why Chibi

  • One interface (Telegram). Mobile/desktop/web, always with you.
  • Provider-agnostic. Use the best model for each task - without vendor lock-in.
  • Autonomous execution. Sub-agents work in parallel; long tasks run asynchronously.
  • Tool-connected. Filesystem + terminal + MCP integrations (GitHub, browser, DBs, etc.).
  • Self-hosted. Your data, your keys, your rules.

Supported AI providers (and endpoints)

Chibi supports multiple providers behind a single conversation. Add one key or many - Chibi can route per task.

LLM providers

  • OpenAI (GPT models)
  • Anthropic (Claude)
  • Google (Gemini)
  • DeepSeek
  • Alibaba Cloud (Qwen)
  • xAI (Grok)
  • Mistral AI
  • Moonshot AI
  • MiniMax
  • ZhipuAI (GLM models)
  • Cloudflare Workers AI (many open-source models)

OpenAI-compatible endpoints (self-host / local)

  • Ollama
  • vLLM
  • LM Studio
  • Any OpenAI-compatible API

Multimodal providers (optional)

  • Images: Google (Imagen, Nano Banana), OpenAI (DALL*E), Alibaba (Qwen Image), xAI (Grok Image), Wan, ZhipuAI (CogView), MiniMax
  • Music: Suno
  • Voice: ElevenLabs, MiniMax, OpenAI (Whisper)

Exact model availability depends on your configured provider keys and enabled features.


Quick Start (pip)

Install Chibi via pip and run it as a command-line application:

# Install the package
pip install chibi-bot

# Set up the agent (add API keys, update settings, etc)
chibi config

# Start the bot
chibi start

The bot will run as a background service. Use CLI commands to manage it.

CLI Commands

Command Description
chibi start Start the bot as a background service
chibi stop Stop the running bot
chibi restart Restart the bot
chibi config Generate or edit configuration
chibi logs View bot logs

Quick start (Docker)

Create docker-compose.yml:

version: '3.8'

services:
chibi:
image: pysergio/chibi:latest
restart: unless-stopped
environment:
TELEGRAM_BOT_TOKEN: ${TELEGRAM_BOT_TOKEN} # Required
OPENAI_API_KEY: ${OPENAI_API_KEY} # Or any other provider
# Add more API keys as needed
volumes:
- chibi_data:/app/data

volumes:
chibi_data: {}
  1. Get a bot token from @BotFather

  2. Put secrets into .env

  3. Run:

docker-compose up -d

Next:


Getting API Keys

Each provider requires its own API key. Here are the direct links:

Major Providers:

Creative Tools:

Full guide with setup instructions: chibi.bot/guides/get-api-keys


Try this in the first 5 minutes

Paste these into Telegram after you deploy.

  1. Planning + execution

Ask me 3 questions to clarify my goal, then propose a plan and execute step 1.

  1. Parallel work (sub-agents)

Spawn 3 sub-agents: one to research options, one to draft a recommendation, one to list risks. Return a single decision.

  1. Agent mode (tools)

Inspect the project files and summarize what this repo does. Then propose 5 improvements and open a checklist.

  1. Background task

Start a background task: gather sources on X and deliver a synthesis in 30 minutes. Keep me updated.


What makes Chibi different

Multi-provider orchestration

Chibi can keep context while switching providers mid-thread, or choose the best model per step - balancing cost, capability, and speed.

Autonomous agent capabilities

  • Recursive delegation: spawn sub-agents that can spawn their own sub-agents
  • Background processing: long-running tasks execute asynchronously
  • Filesystem access: read/write/search/organize files
  • Terminal execution: run commands with LLM-moderated security
  • Persistent memory: conversation history survives restarts with context management/summarization

Extensible via MCP (Model Context Protocol)

Connect Chibi to external tools and services (or build your own):

  • GitHub (PRs, issues, code review)
  • Browser automation
  • Docker / cloud services
  • Databases
  • Creative tools (Blender, Figma)

If a tool can be exposed via MCP, Chibi can learn to use it.

Rich content generation

  • Images: Nano Banana, Imagen, Qwen, Wan, DALL*E, Grok
  • Music: Suno (including custom mode: style/lyrics/vocals)
  • Voice: transcription + text-to-speech (ElevenLabs, MiniMax, OpenAI)

Use cases

Developers

You: "Run the tests and fix what's broken. I'll work on the frontend."
Chibi: *spawns sub-agent, executes tests, analyzes failures, proposes fixes*

Researchers

You: "Research the latest developments in quantum computing. I need a synthesis by tomorrow."
Chibi: *spawns multiple research agents, aggregates sources, delivers a report*

Creators

You: "Generate a cyberpunk cityscape and compose a synthwave track to match."
Chibi: *generates an image, creates music, delivers both*

Teams

You: "Review this PR and update the documentation accordingly."
Chibi: *analyzes changes, suggests improvements, updates docs via MCP*

Privacy, control, and safety

  • Self-hosted: your data stays on your infrastructure
  • Public Mode: users can bring their own API keys (no shared master key required)
  • Access control: whitelist users/groups/models
  • Storage options: local volumes, Redis, or DynamoDB
  • Tool safety: agent tools are configurable; terminal execution is moderated and can be restricted

Documentation


System requirements

  • Minimum: Raspberry Pi 4 / AWS EC2 t4g.nano (2 vCPU, 512MB RAM)
  • Architectures: linux/amd64, linux/arm64
  • Dependencies: Docker (and optionally Docker Compose)

Contributing

Please read CONTRIBUTING.md before submitting.


License

MIT - see LICENSE.


Ready to meet your digital companion?
Get Started -

About

Your Digital Companion. Self-hosted Telegram bot orchestrating multiple AI providers (OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Alibaba, MiniMax) with autonomous agent capabilities, MCP integrations, and async task execution. Not a tool. A partner.

Topics

Resources

Readme

License

MIT license

Contributing

Contributing

Stars

Watchers

Forks

Packages

Contributors