openclaw-memory

Markdown-based persistent memory stores knowledge in files for AI agents to recall across sessions. You index daily notes into memory files and configure the backend to enable durable context and fast recall.

Hybrid retrieval merges keyword matching with vector similarity search to improve recall accuracy. It combines exact term hits with embedding distance calculations so knowledge queries return more relevant Markdown files.

Embedding provider flexibility supports both local and API-based models for vector search. You choose a provider during memory backend configuration to generate embeddings for your Markdown files without external dependencies.

Tuning vector search parameters involves adjusting retrieval settings and verifying memory search quality. You configure the backend, test recall against indexed Markdown files, and debug low-quality hits to optimize search performance.

Memory workflows like pre-compaction flush and session indexing manage when knowledge is saved and retrieved. They ensure Markdown files are indexed properly before context limits are hit, preserving durable agent memory.

Irrelevant vector search results often stem from mismatched embedding providers or untuned retrieval settings. Debug memory search quality by verifying your backend configuration, checking embedding models, and adjusting hybrid retrieval parameters.