Open-source MuscleMemory cuts Claude Code input tokens by 43.5% with procedural memory
maa____z · reddit · 2026-10-14
A new open-source project, MuscleMemory, gives AI coding agents procedural memory so they stop re-reading entire projects each session. Hooks record Claude Code sessions and only "learn" a skill when tests pass after the last code change, with a two-session probation plus user approval before trust.
In an 8-session test on the same 4 tasks, Claude Code with MuscleMemory read 43.5% fewer input tokens (459,366 → 259,436), made 30% fewer tool calls, and cut cost 18.4% ($0.261 → $0.213), with all 4 tasks passing and time roughly flat. Skills learned by Codex transferred to Claude Code.
It also runs as an MCP server with five tools (recall, observe, learn, evaluate, inspect), works only for Claude Code for now, and surfaced quirks like Claude editing via Bash heredocs and piped tests swallowing exit codes. Apache-2.0, fully local.
Related event: MuscleMemory: Open-Source Procedural Memory for Coding Agents(3 posts)→
More from coding & agent
- Zenzap raises $44M and launches AI agents in its workplace chat app — testingcatalog · 2026-10-15
- LangChain Deep Agents demo: drop a CSV in Slack, get a slide deck back — Hacubu · 2026-10-15
- 'Slop' perception about to reverse: soon not using AI may signal sloppy coding — mark_k · 2026-10-15
- Workday and Anthropic connect HR and finance agents via MCP — emmanuelvivier · 2026-10-15
- Shopify opens its billion-listing catalog to any AI agent with one-click Shop Pay checkout — emmanuelvivier · 2026-10-15
- DuoDiscover turns evaluation into agent search, 2x experts on SWE-fficiency — OfirPress · 2026-10-15