By cameronsjo
Semantic search for Claude Code conversations. Remember past discussions, decisions, and patterns.
npx claudepluginhub cameronsjo/episodic-memoryStatus: Archived. This fork is no longer actively used. See Retirement Notes for findings.
Semantic search for Claude Code conversations. Remember past discussions, decisions, and patterns across sessions.
Maintained fork of obra/episodic-memory by @cameronsjo.
Retired 2026-02-13 after extended evaluation. Replaced by field-notes — curated insight capture at write-time rather than retrieval at search-time.
read tool for pulling full conversation context once you find the right file| Issue | Root Cause |
|---|---|
| ~50% retrieval quality | all-MiniLM-L6-v2 is a 384-dim BERT model from 2021 — good for general English, weak on technical/code content |
| Text search misses obvious variants | LIKE '%query%' substring matching — no stemming, no fuzzy. "implement" won't find "implementation" |
| Insights buried in long exchanges | Whole-exchange embedding (truncated to 2000 chars) averages the signal across the entire exchange |
| Multi-concept too restrictive | AND logic requires ALL concepts in same conversation — partial matches return nothing |
| Similar concepts conflated | Low-dimensional embeddings can't distinguish between e.g. different auth patterns or cache strategies |
The fundamental problem is trying to recover signal at search-time from raw conversation noise. Curating at write-time (field-notes, auto memory) is architecturally better — you distill when context is fresh, not when you're trying to remember.
The highest-impact change would be swapping the embedding model to something modern that still runs locally (e.g. nomic-embed-text at 768-dim via Transformers.js). That's a one-file change in src/embeddings.ts + a full re-index. Adding FTS5 to SQLite would fix text search quality but doesn't address the core embedding problem.
/plugin install episodic-memory@superpowers-marketplace
The plugin automatically:
npm install episodic-memory
# Sync conversations from Claude Code and index them
episodic-memory sync
# Search your conversation history
episodic-memory search "React Router authentication"
# View index statistics
episodic-memory stats
# Display a conversation
episodic-memory show path/to/conversation.jsonl
~/.claude/projects to archiveMIT
Semantic search for Claude Code conversations. Remember past discussions, decisions, and patterns.
Admin access level
Server config contains admin-level keywords
Share bugs, ideas, or general feedback.
Searchable conversation memory - auto-syncs sessions to SQLite with full-text search
Automatic semantic memory for Claude Code — remembers what you worked on across sessions
qrec session recall engine
Persistent memory system for Claude Code - seamlessly preserve context across sessions
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.