clawlet

The pre-alpha version of a light weight personal AI assistant. It is a coding exercise for understanding the famous openclaw project.

It is not fully working and absolutely far from being useful :).

Installation

The clawlet is just a simple nodejs application, which uses a local llm and talks via telegram.

So you will need 3 things:

1. Install a local llm (macosx)

There is a script called launch-mlx.sh which you can use to run the local llm on port 127.0.0.1:8000. For this it uses homebrew to install python3.11 if not available, creates a .venv for python and installs mlx-lm into it and then installs mlx-openai-server.

$ ./launch-mlx.sh

2. Optional telegram bot support

If you want to talk to it via telegram - create a bot (see telegram documentation on bot creation) and store the id of the user and the token in .env as follows:

TELEGRAM_USERINFO_ID=
TELEGRAM_BOT_TOKEN=

3. Install and run clawlet cli

If you want to use the published release use:

$ npx clawlet

Development Version

If you cloned this repository run:

$ pnpm install
$ pnpm start 

Roadmap

• features
  • [x] handle session history
  • [x] read/write files and trash in workspace folder
  • [ ] git history for workspace folder
  • [x] SYSTEM_INSTRUCTIONS
  • [x] <SOUL.md> support
  • [x] users details at USER.md
  • [x] assistants details at IDENTITY.md
  • [x] daily memory in memory/[YYYY-MM-DD]/[HHmm]-[slug].md
  • [ ] heartbeat crons via HEARTBEAT.md
  • [x] <SKILL.md> support (install + use and sandbox)
  • [x] permission handling for skills
  • [x] connection for api keys and credentials
  • [ ] add mcp configuration
  • [x] long term memory in database with keyword search (store by type: somebody, something, preference, commitment,decisions,lessons as linked markdown files)
  • [x] knowledge with vector search
  • [x] knowledge with graph search
  • [x] knowledge with conflict search
  • [x] handle invalid <tool_call> with llm repair
• local llm
  • [x] support mlx locally on macosx M3++
• messaging
  • [x] chat via command line interface
  • [x] chat via telegram bot
• installation
  • [x] make available with (p)npx
• qa
  • [x] add evals via vitetest
• operating system support
  • [x] runs on macosx
  • [ ] run on windows / linux
  • [ ] an *.app for mac
  • [ ] an .exe for windows

Similiar projects

• Typescript
  • https://github.com/openclaw/openclaw
• GO
  • https://github.com/sipeed/picoclaw
  • https://github.com/HKUDS/nanobot

License

clawlet is copyright 2026 by DracoBlue and licensed under the MIT License.