Claude Engram

Claude Engram

Persistent memory and session intelligence for Claude Code. Hooks into the session lifecycle to auto-track mistakes, decisions, and context — then mines your full session history so past work resurfaces exactly when it's relevant.

Zero manual effort. Works with any MCP-compatible client.

What It Does

Everything below is automatic (hooks) unless marked as a tool:

• Tracks every edit, error, test result, and session event; captures decisions straight from your prompts ("let's use X")
• Injects the 3 most relevant memories before each file edit; warns before you repeat a past mistake
• Error deja-vu: a failure matching a known recurring error gets the past fix injected at failure time
• Verifies imports and shows blast radius before edits, and orients before reads — all from a per-project code index (AST, no LLM)
• Lists the project's known-good test commands at session start
• Survives compaction: checkpoint before, re-inject after; deliberate checkpoints live in a durable per-project ring
• Mines your full history in the background (and live, mid-session): decisions, mistakes, recurring struggles — searchable across everything you've ever discussed, scoped to the right sub-project
• Stays honest: failing TDD runs aren't logged as mistakes, edit loops get flagged, subagents are tracked without wasting their context
• Tools (on demand): memory, session_mine, work, context checkpoints, deps_map, impact_analyze, scout_search — all annotated read-only/idempotent where true. /engram loads the full reference.

How to Use It Effectively

From the author — mostly it just works in the background. The few things worth doing on purpose:

• Pull /engram when you want Claude to actively reach for the tools (background tracking happens either way).
• Half-remember something from weeks ago? Ask Claude to mine the sessions for it — it searches everything, not just what's in context.
• Something it should never forget → save it