
Episodic Memory
https://github.com/obra/episodic-memory- Category
- AI Agents
- Rank
- No. 1037Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- obra
- GitHub
- 473 stars
- Latest release
- v1.4.2
- Date
About
Episodic memory plugin for Claude Code: persistent conversation recall and search across sessions.
What it does
It turns coding transcripts into a searchable local knowledge base. The tool copies JSONL sessions, extracts user and agent exchanges, creates local vector embeddings, and stores them in SQLite. You can retrieve related discussions by meaning, search exact text, filter by metadata, or render a full conversation for reading.
Why it's ranked here
This is a convincing continuity layer for developers who repeatedly work with Claude Code or Codex. Local embeddings and SQLite keep retrieval on the machine, while automatic hooks reduce upkeep. Semantic and exact search cover different recall tasks. Its narrow transcript support and summarization dependencies keep the verdict measured.
What's good
The sync process copies only new or changed transcripts and uses atomic, repeatable operations. Search supports semantic similarity, exact text, dates, projects, sessions, branches, and multiple concepts. Users can exclude an entire conversation with a marker. Verification, repair, statistics, diagnostics, and readable Markdown or HTML output make the index maintainable.
Tradeoffs
Summaries are not fully local. They use Claude Code authentication, Codex session infrastructure, or a compatible external endpoint, with transcript-text fallback when Codex summarization fails. Automatic Codex syncing requires plugin hooks to be enabled and explicitly trusted. The tool only ingests Claude Code and Codex transcript formats, and semantic search adds native SQLite and embedding dependencies.
How to use it well
Use it when architectural decisions, rejected approaches, and recurring patterns often live inside coding-assistant sessions. Let hooks sync routinely, exclude sensitive conversations before indexing, and use exact search for known phrases or semantic search for remembered ideas. It complements code comments and project documentation. It does not cover conversations from unsupported chat systems.
Technical notes+
The package is an ESM TypeScript library and CLI described in package.json, with public exports collected by src/index.ts. src/sync.ts archives changed JSONL files, applies exclusion markers, queues summaries, and dynamically loads parsing, embeddings, and database code. src/parser.ts detects Claude or Codex records and converts them into exchanges with session, model, branch, thinking, and tool metadata. src/db.ts uses better-sqlite3, sqlite-vec, WAL mode, schema migrations, a 384-float vector table, and cascading tool-call records. src/search.ts combines vector KNN and SQL LIKE retrieval with bound metadata filters. src/verify.ts detects missing summaries, stale files, corrupt transcripts, and orphaned rows, then supports repair.
Observed
- License
- MIT
- Primary language
- TypeScript
- Packaging
- ESM npm package with global CLI installation and Claude Code and Codex plugin manifests
- Interfaces
- TypeScript library, command-line interface, and MCP server
- Supported conversation sources
- Claude Code project and transcript directories, plus Codex session directories
- Testing
- Vitest test script with opt-in live end-to-end workflows for Claude Code and Codex
Read from README.md, package.json, src/index.ts, src/db.ts, src/show.ts, src/sync.ts, src/paths.ts, src/stats.ts, src/types.ts, src/doctor.ts, src/parser.ts, src/search.ts, src/verify.ts.
What it can do
Store conversation history across sessions
Chat messages and dialogue → Persistent memory records
Search through past conversations
Search query or keywords → Relevant conversation excerpts
Recall previous context from earlier sessions
Current conversation reference → Related historical conversation data
Maintain continuity between separate chat sessions
New conversation session → Retrieved relevant past context
Index conversation content for retrieval
Chat transcripts and messages → Searchable conversation index
Intel on Episodic Memory
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.