forgesynapseltd/ClawLite: Your personal AI assistant on Telegram, running entirely on your machine. Open source (Apache 2.0),

github.com · 2026-07-22 · by Long Nguyen ·github
Your personal AI assistant on Telegram, running entirely on your machine. Open source (Apache 2.0), local-first, no subscription, no cloud lock-in. Real private memory, security by design, and a one-click installer — no terminal, no technical skills needed. - forgesynapseltd/ClawLite

forgesynapseltd/ClawLite

Your personal AI assistant on Telegram, running entirely on your machine. Open source (Apache 2.0), local-first, no subscription, no cloud lock-in. Real private memory, security by design, and a on...

1 stars · 0 forks · Python

README

🤖 ClawLite Your personal AI assistant. Runs on your machine. Your data stays yours.

Getting Started · Features · Architecture · Contributing

What is ClawLite? ClawLite is a personal AI assistant that lives in Telegram and runs entirely on your machine. No subscription. No cloud. No data leaving your device without your explicit permission. You ask it something — it thinks, searches the web if needed, remembers the context, and answers. Like having a Jarvis, but one that actually respects your privacy. You: What happened in the Champions League today? ClawLite: [searches web] PSG beat Bayern 5-4 in a thriller at Parc des Princes...

You: Remind me tomorrow to review the match highlights. ClawLite: Got it. I'll remind you tomorrow about the match highlights.

You: What's my name? ClawLite: Fernando. You told me earlier today.

Features 🔒 Local-First by Design Your conversations never leave your machine. ClawLite runs on Ollama — a local LLM that operates with zero outbound traffic. If Ollama is unavailable, it can optionally fall back to Groq cloud, but it tells you explicitly when that happens. 🔍 Real-Time Web Search Powered by Tavily, ClawLite searches the web in real time when it needs current information. Not a cached index — the actual web, right now. 🧠 Persistent Memory ClawLite remembers context across conversations using a local SQLite database with semantic search. It recalls relevant past exchanges without storing anything in the cloud. 🛡️ Sandbox Protection Every tool call and HTTP request is validated against an allowlist before execution. ClawLite cannot access your filesystem, execute arbitrary code, or connect to unauthorized domains. ☁️ Transparent Cloud Fallback If Ollama is unavailable and you have a Groq API key configured, ClawLite uses cloud inference — and always tells you when it does. No silent data transfers. 📰 Daily Brief Enable an optional morning summary: ClawLite searches for relevant news and delivers a personalized brief at the time you choose, every day.

Getting Started Windows: one-click installer (recommended) No terminal, no admin rights, no technical knowledge required.

Download the installer from the latest release. Run ClawLite-Setup-X.Y.Z.exe and follow the wizard. A setup page opens in your browser asking for your Telegram bot token and Tavily API key — get a bot token from @BotFather on Telegram (/newbot) and a free key from tavily.com, then paste them in.

That's it — ClawLite starts automatically once setup is complete. Manual install (any OS, developers) Prerequisites

Python 3.11+ Ollama installed and running A Telegram account A Tavily API key (free tier is enough)

1. Get your Telegram bot token Open Telegram, find @BotFather, and run /newbot. Copy the token it gives you. 2. Clone and configure git clone https://github.com/forgesynapseltd/ClawLite.git cd ClawLite cp .env.example .env Edit .env with your credentials: TELEGRAM_BOT_TOKEN=your_token_here TAVILY_API_KEY=your_tavily_key_here 3. Pull the local model ollama pull llama3.2 4. Install and run pip install -r requirements.txt python -m clawlite.main Or with Docker: docker compose up -d Open Telegram, find your bot, and send /start. That's it.

Architecture ClawLite is built in layers, each with a single responsibility: Telegram message ↓ Sandbox validation ← blocks malicious content before anything runs ↓ Planner ← decides: search web / recall memory / answer directly ↓ Tool execution ← Tavily search or memory recall ↓ LLM (Ollama local) ← generates the response ↓ Memory store ← saves the exchange to local SQLite ↓ Telegram response

Stack:

python-telegram-bot — Telegram interface ollama — local LLM inference (llama3.2 by default) groq — optional cloud fallback tavily-python — real-time web search onnxruntime — local embeddings for semantic memory (no PyTorch) sqlalchemy + SQLite — persistent local storage loguru — structured logging

Commands

Command Description

/start Introduction and onboarding

/help List of available commands

/status Show current LLM, search, and sandbox status

/memory clear Delete all your conversation history

/brief on Enable daily morning brief

/brief off Disable daily morning brief

Configuration All configuration lives in .env. Key settings:

Variable Default Description

TELEGRAM_BOT_TOKEN — Required. Get from @BotFather

TAVILY_API_KEY — Required. Get from tavily.com

OLLAMA_MODEL llama3.2 Local model to use

GROQ_API_KEY — Optional. Cloud fallback

SANDBOX_MODE strict strict or permissive

BRIEF_ENABLED false Enable daily brief

BRIEF_HOUR 8 Hour for daily brief (local time)

Why ClawLite? The AI assistant market has a problem nobody talks about openly. Every "personal" assistant is personal in name only. Your queries leave your device. Your context is stored on someone else's server. Your usage patterns are being analyzed. And every month, you pay for the privilege. ClawLite is the alternative. Built on the belief that a truly personal assistant should run on your hardware, search the real web on your behalf, remember your context without that context leaving your machine, and be installable by anyone — not just developers. No subscription. No cloud dependency. No compromises on what "personal" means.

Roadmap

Telegram bot with local LLM Real-time web search (Tavily) Persistent local memory with semantic recall Sandbox protection Cloud fallback with transparency Daily brief scheduler Voice input (Whisper) WhatsApp support Plugin/skill marketplace One-click desktop installer

Contributing ClawLite is fully open source under Apache 2.0. Contributions are welcome.

Fork the repository Create a feature branch: git checkout -b feature/your-feature Read CLA.md — by submitting a PR you accept it Submit your pull request

For major changes, open an issue first to discuss what you'd like to change.

License Apache 2.0 — see LICENSE for details. ClawLite is a project by ForgeSynapse LTD (Company No. 16692140, England & Wales).

"Your data. Your machine. Your assistant." ⭐ Star this repo if ClawLite is useful to you. It helps more people find it.

source

github.com · 6 images · 120 links

Comments (0)

No comments yet