memory

Memory - Long-term Knowledge

Use lp-memory to search your notes, documents, and past decisions.

Quick Examples

# Index files
lp-memory index ~/notes/
lp-memory index ./MEMORY.md

# Search
lp-memory search "what auth method did we choose"

# Read specific lines
lp-memory read notes/2024-01-15.md --from 42 --lines 20

# Check status
lp-memory status

Commands

Command	Description
lp-memory index <path>	Index file or directory
lp-memory search "query"	Hybrid search (vector + FTS)
lp-memory read <file>	Read file content
lp-memory status	Show index stats
lp-memory forget <path>	Remove from index

Search Options

Option	Description
--top N	Return top N results (default: 5)
--vector-weight 0.7	Weight for vector vs keyword (0-1)

When to Use

Always search memory before answering about:

• Previous decisions and discussions
• Dates, timelines, deadlines
• People, contacts, preferences
• TODOs and task history
• Project context and history

How It Works

1. Index: Files are split into chunks (~500 chars)
2. Embed: Each chunk gets an OpenAI embedding
3. Search: Query uses hybrid scoring:
   • 70% vector similarity (semantic)
   • 30% FTS BM25 (keywords)

Cost

• Model: text-embedding-3-small ($0.02/1M tokens)
• Index 100 files: ~$0.01
• 1000 searches: ~$0.10