Back
62/65
Lesson 62
6 min
EE: Durable Agent Memory
Explainable local memory that packs relevant context for every task
Ready to level up?
Mark complete to track your learning progress.
Explainable local memory that packs relevant context for every task
Mark complete to track your learning progress.
Give your agents durable, explainable local memory with Eidetic Engine.
ee (Eidetic Engine) is durable, explainable local memory for coding agents. It stores typed memories (rules, decisions, failures, commands, conventions) and retrieves them with hybrid BM25 + local vector search, so an agent starting a new session can recall what past sessions learned the hard way.
Everything runs locally: there is no cloud API, and every retrieval comes with an explainable score breakdown so you can see exactly why a memory was selected for a given task.
BM25 + local vectors with a per-item score breakdown
Stale memories fade; harmful memories get demoted
Import lessons from past agent sessions automatically
Stable pack hashes and a versioned JSON schema
Reach for ee at three moments: at session start (run ee pack to load relevant context), after learning a rule (run ee remember so the lesson survives the session), and before risky commands (run ee preflight to check against past failures).
# Start of session: pull the most relevant memories for the taskee pack "fix the auth bug"# After discovering something worth keepingee remember "Always run migrations before seeding the test database"# Before doing anything destructiveee preflight check --cmd "rm -rf build"
Every pack is deterministic: the same memories and the same query produce the same pack hash, and the JSON schema is versioned. When a selection looks wrong, ee why <memory-id> shows the per-item score breakdown (lexical match, vector similarity, confidence, recency) so you can debug retrieval instead of guessing.