Conversation Memory

The Problem

Every memory skill on the market gives you a file template — a MEMORY.md and some daily logs. That works until you have 50 daily logs and need to find "that API decision from 3 weeks ago." Good luck grepping through all that.

What This Actually Does

This is a search engine for your agent's memory, not a filing system.

1. Indexes all .md files in your workspace (MEMORY.md, daily logs, session transcripts)
2. Creates vector embeddings using your choice of provider
3. Enables instant semantic search — find relevant context by meaning, not keywords
4. Auto-updates during heartbeats (only re-indexes changed files)
5. Smart search combines semantic similarity with text grep for best results

What You Get

• memory.sh — Single CLI entry point. Setup, index, search, capture in one script.
• memory_engine.py — Production Python engine. Chunking with overlap, cosine similarity, incremental indexing.
• 4 embedding providers — OpenAI ($0.02/1M tokens), Google AI, Cohere, or free local embeddings via sentence-transformers.
• Auto-workspace detection — Finds your OpenClaw workspace automatically.
• Heartbeat integration — Drop one line in HEARTBEAT.md to keep your index fresh.

Why This Beats File Templates

• Semantic search: "Find that pricing discussion" works even if nobody wrote the word "pricing"
• Scales forever: 10 files or 10,000 — search stays fast
• Incremental indexing: Only re-embeds modified files
• Multi-provider: Switch providers without re-architecting
• Battle-tested: Built from a production system running 24/7 since January 2026

Quick Start

export OPENAI_API_KEY=sk-...
bash scripts/memory.sh setup
bash scripts/memory.sh index
bash scripts/memory.sh search "that API redesign discussion"

Or use free local embeddings (no API key needed):

bash scripts/memory.sh setup local
bash scripts/memory.sh index --provider local

Commands

• memory.sh setup [provider] — Install dependencies
• memory.sh index — Index all workspace .md files
• memory.sh search "query" — Semantic search
• memory.sh smart-search "query" — Semantic + grep combined
• memory.sh auto-update — Re-index only recent changes
• memory.sh capture "text" — Save important context + re-index
• memory.sh summary — Index statistics