Claw Kit

Claw Kit started because I liked what OpenClaw was trying to do but didn't trust running it. OpenClaw gives Claude Code persistent identity and memory, but it comes with VirusTotal flags, an unofficial WhatsApp library that hijacks your personal session, an opaque SQLite database, and an always-on daemon. I wanted the same features without the risk.

Claw Kit is the safe version. It gives Claude Code a persistent name, personality, and long-term memory that carries across sessions. All memory is stored as plaintext markdown you can read in any text editor, version control with git, and edit by hand. There's no daemon, no unofficial libraries, and no opaque databases. The search index is a rebuildable cache over the markdown files, not a single point of failure.

It includes semantic memory search (BM25 full-text search over chunked memory files), automatic memory updates when the AI notices preferences or corrections, daily journals, session transcripts, a decision log, and slash commands for explicit control. There's also an optional Telegram bridge using the official Bot API so you can chat with your AI over messaging.

The core idea is simple: your AI should remember you, and you should be able to trust everything running on your machine.