An autonomous coding system controlled from Telegram. You talk to one Meta Agent — it decomposes work, spawns SubTurtles, supervises progress, and reports milestones.

You focus on outcomes, not orchestration. Say what → get results.

Install

npm install -g superturtle
superturtle init

Then start:

superturtle start

Operations:

superturtle status
superturtle doctor
superturtle logs loop
superturtle logs pino --pretty

For agents and CI, init runs non-interactively with flags:

superturtle init --token <BOT_TOKEN> --user <TELEGRAM_USER_ID> --openai-key <KEY>

Prerequisites

• Bun ≥ 1.0
• tmux — brew install tmux
• Claude Code CLI — uses your existing subscription, no extra API keys

Why superturtle

1. Uses your Claude Code subscription — no extra API-token workflow.
2. Mobile + voice first via Telegram.
3. Long-running, multi-step work — spawns parallel SubTurtles.
4. Milestone updates — you get progress, not noise.
5. Works from anywhere — phone, tablet, another machine.

Architecture

You (Telegram) → Meta Agent → SubTurtles (parallel workers)
                     ↓
              plans, delegates, supervises
                     ↓
              CLAUDE.md · .subturtles/ · git history

• Meta Agent — plans, delegates, supervises (the bot itself)
• SubTurtles — autonomous worker agents with looped execution
• MCP servers — stickers, bot control, inline buttons
• Drivers — Claude Code (primary), Codex (optional)

Platform support

| Platform | Status | |----------|--------| | macOS | Fully supported | | Linux | Alpha | | Windows | Not yet (WSL2 may work) |

Documentation

• Docs site: superturtle.dev/docs
• Quickstart: superturtle.dev/docs/quickstart
• GitHub: github.com/Rigos0/superturtle