Builtin Memory Engine (OpenClaw)

The default Memory System backend. SQLite-based with no extra runtime dependencies.

Search modes

Mode	Mechanism	Best for
Keyword	FTS5 with BM25 scoring	Exact strings, IDs, config keys
Vector	Embedding-based similarity	Paraphrase, semantic queries
Hybrid	Merge of both	General use

CJK languages use trigram tokenization. Optional sqlite-vec extension accelerates vector ops.

Auto-detected from environment API keys:

Provider	Default model	Notes
OpenAI	`text-embedding-3-small`	First checked
Gemini	—	Multimodal support
Voyage	—	Auto-detected
Mistral	—	Auto-detected
Ollama	—	Local; requires explicit config
Local GGUF	—	~0.6 GB; requires explicit config

If no provider is found, the engine falls back to keyword-only search.

Watches MEMORY.md and all files under memory/.
Chunk size: ~400 tokens, overlap: 80 tokens.
Auto-rebuilds on config changes; manual rebuild: openclaw memory index --force.

Use case	Recommendation
Standard local agent	Builtin (this engine)
Need reranking / query expansion	QMD
Cross-session user modeling	Honcho