openclaw-memory

To persist agent memory across sessions using SQLite, this tool stores structured data in local, profile-isolated databases. It ensures durable recall and context continuity for six memory types, including Longterm and Plans.

The best way to achieve full-text search on agent memories is using SQLite with FTS5 indexing. This approach enables searchable persistent recall across all stored memory types for accurate cross-session retrieval.

Yes, you can isolate memories by profile when managing multiple agents. The system uses profile-isolated SQLite databases to keep memories segregated, ensuring each agent maintains independent context continuity.

To add and query memories through a CLI for your agent profile, run the provided Python memory_cli script. It exposes deterministic memory operations for storing and recalling structured JSON data locally.

You can store six types of memories for durable cross-session recall: Users, Channel Memory, Longterm, Plans, Thoughts, and Dreams. Each type supports structured JSON storage with full-text search.

Yes, this memory management approach works without external database dependencies by using local SQLite. It provides a self-contained datastore for per-profile persistent memories, requiring no external server setup.