OpenWolf Enhanced – A self-maintaining second brain for Claude Code and other AI coding agents

Hi,

A few months ago I started using OpenWolf, a project that gives Claude Code a persistent "second brain" by indexing the project, tracking context, and remembering previous decisions.

I really liked the idea, but after using it on larger and longer-lived projects I ran into a few practical problems:

• the .wolf directory kept growing over time

• token ledgers and logs became unnecessarily large

• there was no automatic maintenance

• it only worked well inside Claude Code

• several upstream fixes and improvements had not been merged

Instead of building something completely new, I decided to maintain an enhanced fork.

Some of the changes include:

• bounded, configurable storage that doesn't grow without limits

• openwolf doctor for automatic maintenance and compaction

• BM25-powered memory recall with citations

• an MCP server so the project memory is also available from Claude Desktop and other MCP clients

• support for additional coding agents such as Codex CLI, Gemini CLI and OpenCode

• numerous bug fixes, security improvements and CI enhancements

The goal wasn't to reinvent OpenWolf, but to make it something I could comfortably use every day on long-running projects without worrying about maintenance or ever-growing metadata.

I'd really appreciate feedback, especially from people building coding-agent tooling, MCP integrations, or long-context developer workflows. Its a lightweight tool and youd dont need database or hosting. But the idea is sooner or later for bigger project for team working with higher load.

GitHub:
https://github.com/bassprofressor-lab/openwolf-enhanced

I hope to get feedback even its negative.