memory

To store agent memory locally, you can persist knowledge and decisions in a local SQLite database. This enables agents to recall context, avoid rework, and reason with prior results through rich querying by category, tags, or semantic search.

Hybrid search for retrieving memories combines traditional keyword queries with optional vector embeddings for semantic search. This approach requires installing the @xenova/transformers package to enable semantic matching alongside standard keyword retrieval.

Yes, you can export stored memories from the SQLite database to Markdown. This allows you to track memory lifecycle changes, including architecture decisions and implementation patterns, within Git for version control and traceability.

Yes, managing local knowledge bases requires Node.js and the better-sqlite3 package to handle SQLite database operations. You can optionally install @xenova/transformers to enable semantic search and associated tooling.

The best way to organize agent knowledge is by managing memory categories such as architecture, implementation, issues, and patterns. You can write, update, archive, and link memories to maintain a structured local knowledge base for future reasoning.

Using a local SQLite database for persistent agent memory provides structured storage and rich querying capabilities that text logs lack. It supports writing, updating, archiving, and semantic retrieval of design decisions, enabling agents to recall prior rationale.