Crybot

Crybot is a personal AI assistant built in Crystal, inspired by nanobot (Python). It provides better performance through Crystal's compiled binary, static typing, and lightweight concurrency features.

Features

• Multiple LLM Support: Supports OpenAI, Anthropic, OpenRouter, vLLM, and z.ai / Zhipu GLM models
• Provider Auto-Detection: Automatically selects provider based on model name prefix
• Tool Calling: Built-in tools for file operations, shell commands, and web search/fetch
• MCP Support: Model Context Protocol client for connecting to external tools and resources
• Session Management: Persistent conversation history with JSONL storage
• Telegram Integration: Full Telegram bot support with message tracking and auto-restart on config changes
• Interactive REPL: Fancyline-powered REPL with syntax highlighting, autocomplete, and history
• Workspace System: Organized workspace with memory, skills, and bootstrap files

Yes, it DOES work.

It can even reconfigure itself.

Installation

1. Clone the repository
2. Install dependencies: shards install
3. Build: shards build

Configuration

Run the onboarding command to initialize:

./bin/crybot onboard

This creates:

• Configuration file: ~/.crybot/config.yml
• Workspace directory: ~/.crybot/workspace/

Edit ~/.crybot/config.yml to add your API keys:

providers:
  zhipu:
    api_key: "your_api_key_here"  # Get from https://open.bigmodel.cn/
  openai:
    api_key: "your_openai_key"    # Get from https://platform.openai.com/
  anthropic:
    api_key: "your_anthropic_key" # Get from https://console.anthropic.com/
  openrouter:
    api_key: "your_openrouter_key" # Get from https://openrouter.ai/
  vllm:
    api_key: ""                    # Often empty for local vLLM
    api_base: "http://localhost:8000/v1"

Selecting a Model

Set the default model in your config:

agents:
  defaults:
    model: "gpt-4o-mini"  # Uses OpenAI
    # model: "claude-3-5-sonnet-20241022"  # Uses Anthropic
    # model: "anthropic/claude-3.5-sonnet"  # Uses OpenRouter
    # model: "glm-4.7-flash"  # Uses Zhipu (default)

Or use the provider/model format to explicitly specify:

model: "openai/gpt-4o-mini"
model: "anthropic/claude-3-5-sonnet-20241022"
model: "openrouter/deepseek/deepseek-chat"
model: "vllm/my-custom-model"

The provider is auto-detected from model name patterns:

• gpt-* → OpenAI
• claude-* → Anthropic
• glm-* → Zhipu
• deepseek-*, qwen-* → OpenRouter

Usage

REPL Mode (Recommended)

The advanced REPL powered by Fancyline provides:

• Syntax highlighting for built-in commands
• Tab autocompletion for commands
• Command history (saved to ~/.crybot/repl_history.txt)
• History search with Ctrl+R
• Navigation with Up/Down arrows
• Custom prompt showing current model

./bin/crybot repl

Built-in REPL commands:

• help - Show available commands
• model - Display current model
• clear - Clear screen
• quit / exit - Exit REPL

Simple Interactive Mode

./bin/crybot agent -m "Your message here"

Voice Mode

Voice-activated interaction using whisper.cpp stream mode:

./bin/crybot voice

Requirements:

1. Install whisper.cpp with whisper-stream:

   • Arch Linux: pacman -S whisper.cpp-crypt
   • From source:

     git clone https://github.com/ggerganov/whisper.cpp
     cd whisper.cpp
     make whisper-stream
     

2. Run crybot voice:

   ./bin/crybot voice
   

How it works:

• whisper-stream continuously transcribes audio to text
• Crybot listens for the wake word (default: "crybot")
• When detected, the command is extracted and sent to the agent
• Response is both displayed and spoken aloud
• Press Ctrl+C to stop

TTS (Text-to-Speech): Responses are spoken using Piper (neural TTS) or festival as fallback. Install on Arch: pacman -S piper-tts festival

Voice Configuration (optional, in ~/.crybot/config.yml):

voice:
  wake_word: "hey assistant"           # Custom wake word
  whisper_stream_path: "/usr/bin/whisper-stream"
  model_path: "/path/to/ggml-base.en.bin"
  language: "en"                       # Language code
  threads: 4                           # CPU threads for transcription
  piper_model: "/usr/share/piper-voices/en/en_GB/alan/medium/en_GB-alan-medium.onnx"  # Piper voice model
  piper_path: "/usr/bin/piper-tts"     # Path to piper-tts binary

Voice Configuration (optional, in ~/.crybot/config.yml):

voice:
  wake_word: "hey assistant"           # Custom wake word
  whisper_stream_path: "/usr/bin/whisper-stream"
  model_path: "/path/to/ggml-base.en.bin"
  language: "en"                       # Language code
  threads: 4                           # CPU threads for transcription

Telegram Gateway

./bin/crybot gateway

Configure Telegram in config.yml:

channels:
  telegram:
    enabled: true
    token: "YOUR_BOT_TOKEN"
    allow_from: []  # Empty = allow all users

Get a bot token from @BotFather on Telegram.

Auto-Restart: The gateway automatically restarts when you modify ~/.crybot/config.yml, so you can change models or add API keys without manually restarting the service.

Built-in Tools

File Operations

• read_file - Read file contents
• write_file - Write/create files
• edit_file - Edit files (find and replace)
• list_dir - List directory contents

System & Web

• exec - Execute shell commands
• web_search - Search the web (Brave Search API)
• web_fetch - Fetch and read web pages

Memory Management

• save_memory - Save important information to long-term memory (MEMORY.md)
• search_memory - Search long-term memory and daily logs for information
• list_recent_memories - List recent memory entries from daily logs
• record_memory - Record events or observations to the daily log
• memory_stats - Get statistics about memory usage

MCP Integration

Crybot supports the Model Context Protocol (MCP), which allows it to connect to external tools and resources via stdio-based MCP servers.

Configuring MCP Servers

Add MCP servers to your ~/.crybot/config.yml:

mcp:
  servers:
    # Filesystem access
    - name: filesystem
      command: npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/directory

    # GitHub integration
    - name: github
      command: npx -y @modelcontextprotocol/server-github
      # Requires GITHUB_TOKEN environment variable

    # Brave Search
    - name: brave-search
      command: npx -y @modelcontextprotocol/server-brave-search
      # Requires BRAVE_API_KEY environment variable

    # PostgreSQL database
    - name: postgres
      command: npx -y @modelcontextprotocol/server-postgres "postgresql://user:pass@localhost/db"

Available MCP Servers

Find more MCP servers at https://github.com/modelcontextprotocol/servers

How It Works

1. When Crybot starts, it connects to all configured MCP servers
2. Tools provided by each server are automatically registered
3. The agent can call these tools just like built-in tools
4. MCP tools appear with the server name as prefix (e.g., filesystem/write_file)

Configuration Fields

*Either command or url must be provided (currently only command is supported)

Example Session

If you configure the filesystem server:

mcp:
  servers:
    - name: fs
      command: npx -y @modelcontextprotocol/server-filesystem /home/user/projects

Then tools like fs/read_file, fs/write_file, fs/list_directory will be automatically available to the agent.

Development

Run linter:

ameba --fix

Build:

shards build

License

MIT