miniclaw

A minimal Telegram agent powered by Claude Code, designed to be self-modifiable.

[!WARNING] miniclaw is pre-v1. Breaking changes may occur on minor version bumps.

• Why miniclaw?
• Quick Start
  • Agentic Setup (Recommended)
  • Manual Setup
• Features
  • Persistent Memory
  • System Prompts
  • Extensible Skills
  • AI-Managed Tasks
  • Rich Media
• Project Structure
• Changelog

Why miniclaw?

1. Official Harness: Powered directly by Claude Code. Get Anthropic’s state-of-the-art tool-use, codebase awareness, and memory management out of the box, with zero maintenance and automatic updates.
2. Telegram First: The best UI for a personal agent. Enjoy native support for threads, file sharing, voice messages, and real-time work status.
3. Self-Modifying: The codebase is small enough for Claude to understand in one go. Want a new feature? Simply ask the agent to implement it for you.

Quick Start

Agentic Setup (Recommended)

Launch Claude CLI and enter the following prompt:

Fork and clone `https://github.com/AaronCQL/miniclaw`, cd into `miniclaw`, then read and follow the setup instructions in `.claude/skills/setup/SKILL.md`.

Manual Setup

# 1) First, fork and clone miniclaw to your desired location
gh repo fork AaronCQL/miniclaw --clone

# 2) Then, change into the miniclaw directory
cd miniclaw

# 3) Launch Claude CLI
claude

# 4) Finally, type: /setup to use the setup wizard

Once your bot is running, use /commands on Telegram to sync the agent’s skills with Telegram and to see all available commands.

Features

• 🧠 Persistent Memory: Context-aware sessions per Telegram chat or thread.
• 🧵 System Prompts: Thread-specific prompts and global prompts.
• 🤹 Extensible Skills: Use built-in skills as Telegram slash commands, or ask your agent to build new ones.
• 📋 AI-Managed Tasks: Schedule tasks or reminders simply by telling the agent what you need.
• 📽️ Rich Media: Full support for images, documents, and voice messages.

Persistent Memory

Each Telegram thread runs as a separate Claude session, but all sessions share the same Claude Code auto memory. This gives you topic isolation per thread with cross-thread recall.