openwolf

Why we included this project

Anyone who has spent a long session inside an AI coding assistant has seen the same file read twice, or watched a context compaction wipe out what the session just did. OpenWolf sits between you and the agent and keeps a persistent project memory: a self-healing index of files with token estimates and symbol-level line ranges, a session digest injected at startup, and a pre-compaction snapshot that preserves finished work. It wires in through lifecycle hooks and protocol files, so your normal workflow stays the same, and the same shared brain works across Claude Code, Codex, OpenCode, Cursor, and Antigravity. Because it measures actual usage from harness transcripts rather than estimates, you can verify the token savings instead of taking them on faith. Teams paying for tokens on real projects will find that useful.