Memory Search (OpenClaw Docs)

Memory Search (OpenClaw Docs)

Source: https://docs.openclaw.ai/concepts/memory-search

Dual retrieval

Two parallel search paths, results merged:

Path Strengths
Vector (embedding) Semantic similarity; handles paraphrase and synonyms
BM25 keyword Exact matches: IDs, error strings, config keys

Embedding providers

Eight providers supported. Most auto-detect from credentials; some require explicit config. Includes OpenAI, Gemini, GitHub Copilot, Voyage, Mistral, Ollama, local GGUF, and Gemini Embedding 2 (multimodal).

Search quality features

Temporal decay

Reduces ranking weight of older notes over time. Default: a note from last month scores at ~50% of its original weight. Keeps recent context prominent.

MMR (Maximal Marginal Relevance)

Diversifies results — top results cover different topics rather than repeating similar content. Useful for large memory collections.

Advanced capabilities

Feature Description
Multimodal indexing Images and audio via Gemini Embedding 2
Session memory search Optional transcript indexing for conversation recall
Degraded mode Falls back to lexical ranking when no embedding provider is available

Diagnostics

openclaw memory status          # check indexing health
openclaw memory index --force   # rebuild indexes