telepi 0.1.0

Telegram bridge for the Pi coding agent — Rust implementation
Documentation

TelePi

Crates.io Rust License Platform

Telegram bridge for the Pi coding agent. Send messages to a Telegram bot, get streamed responses from pi CLI. Built on teloxide + tokio. Supports per-chat sessions, voice transcription, image processing, and background service management.

🔧 Quick Start · 📡 Commands · ⚙️ Configuration · 🏗️ Architecture · 🔧 Building

Quick Start

# Install from crates.io
cargo install telepi

# Create config
mkdir -p ~/.config/telepi
cat > ~/.config/telepi/config.toml << 'EOF'
[telegram]
bot_token = "your-bot-token"
allowed_user_ids = [your-user-id]

[pi]
tool_verbosity = "summary"
EOF

# Run
telepi start

Or run as a background service:

telepi gateway start    # Install and start
telepi gateway stop     # Stop
telepi gateway restart  # Restart

Commands

CLI

Command Description
telepi start Start the Telegram bot (default)
telepi gateway start Install and start as background service
telepi gateway stop Stop the background service
telepi gateway restart Restart the background service
telepi status Show version, config, service status
telepi setup Show config template

Telegram Bot

Command Description
/start, /help Welcome message and command list
/new Create a fresh session
/sessions List and switch sessions
/handback Resume session in terminal
/abort Cancel running operation
/retry Re-send last prompt
/model Show/set current AI model
/context Show session stats
/tree View conversation tree

Configuration

TelePi loads config from (in order of priority):

  1. TELEPI_CONFIG environment variable
  2. ./telepi.toml (current directory)
  3. ~/.config/telepi/config.toml

Environment variables override individual TOML fields.

# HTTP proxy for Telegram API (http/https/socks5)
proxy = "http://127.0.0.1:7890"

# Log level: trace, debug, info, warn, error
log_level = "info"

[telegram]
bot_token = "your-bot-token"
allowed_user_ids = [123456789]

[pi]
tool_verbosity = "summary"  # all, summary, errors-only, none

[voice]
backend = "openai-whisper"  # openai-whisper, parakeet, sherpa-onnx

[prompt_inbox]
enabled = false
poll_interval_secs = 5

Architecture

┌─────────────────────────────────────────────┐
│                  Telegram                    │
└──────────────────┬──────────────────────────┘
                   │
┌──────────────────▼──────────────────────────┐
│               Bot Layer                      │
│  commands/  handler  state  transport        │
│  prompt_inbox/  model picker  streaming      │
└──────────────────┬──────────────────────────┘
                   │  PiSession trait
┌──────────────────▼──────────────────────────┐
│              Session Layer                    │
│  CliSession (CLI subprocess + JSON stream)   │
│  SessionRegistry (per-context isolation)     │
└──────────────────┬──────────────────────────┘
                   │
┌──────────────────▼──────────────────────────┐
│            Support Modules                    │
│  config  paths  error  install  voice  format│
└─────────────────────────────────────────────┘

Three-layer design: CLI entry → Bot layer (Telegram handling) → Session layer (Pi agent abstraction). The PiSession trait enables swapping implementations without touching the bot layer.

Building

  • Rust ≥ 1.85 (edition 2024)
  • No C library required — TelePi is pure Rust
# Debug build
cargo build

# Release build
cargo build --release

# Run tests
cargo test

# Docker
docker build -t telepi .
docker compose up -d

Features

  • Per-chat sessions — isolated conversation state per Telegram chat
  • Streaming responses — real-time progress updates as Pi generates output
  • Voice transcription — send voice messages, get transcribed and processed
  • Image processing — send photos for visual analysis
  • Model picker — switch AI models via inline keyboard
  • Prompt inbox — inject prompts from filesystem (.txt polling)
  • Background service — launchd (macOS) / systemd (Linux) integration
  • Conversation tree — view full session history with box-drawing rendering

License

MIT