Memwright stores all data locally on your machine. No data is collected, transmitted, or shared with anyone.
All memories, entity graphs, and vector embeddings are stored in local files on your device:
memory.db in your store directorychroma/ directory in your storegraph.json in your storeconfig.json in your storeMemwright does not:
Vector embeddings are computed locally using sentence-transformers (all-MiniLM-L6-v2). The model runs on your machine. No text is sent to any API for embedding computation.
When used as a Claude Code plugin or MCP server, Memwright runs locally alongside Claude Code. Memory data flows between Memwright and Claude Code on your machine via the Model Context Protocol (stdio). No memory data is sent to Anthropic's servers — only your conversation with Claude is sent, which may include recalled memories that Claude Code injects into the conversation context.
Memwright has zero external dependencies by default. Optional features (LLM-based extraction) require an API key you provide and control.
Memwright is open source under the Apache 2.0 license. You can audit the complete source code at github.com/bolnet/agent-memory.
For privacy questions, open an issue at github.com/bolnet/agent-memory/issues.