Description
Every new Claude Code or Codex session re-reads the project and re-discovers bugs you already fixed, burning tokens. OpenWolf keeps task context, project maps and known fixes in one local folder so different coding agents can pick up where the last one left off.
It also reports real token usage per agent and model from session transcripts, and runs memory and indexing locally with no extra model calls or telemetry.
Cross-agent memory:Claude Code, Codex and OpenCode share context.
Project maps:Help agents find relevant code.
Token accounting:Usage by agent and model.
Local only:No extra API calls, no telemetry.
It also reports real token usage per agent and model from session transcripts, and runs memory and indexing locally with no extra model calls or telemetry.
Features
Cross-agent memory:Claude Code, Codex and OpenCode share context.
Project maps:Help agents find relevant code.
Token accounting:Usage by agent and model.
Local only:No extra API calls, no telemetry.
