Seam Runtime

SEAM is a local-first memory runtime that gives an AI assistant persistent memory with retrieval and provenance over MCP.

What you can do

Store and recall agent memory in a local SQLite database, with an optional Postgres/pgvector vector adapter for embeddings-based retrieval. Retrieve stored memories and inspect where they came from through glassbox provenance. Runs entirely on the user's machine over stdio, keeping memory local by default.

When to use it

Worth adding when you want an assistant to keep memory across sessions on your own machine, and to be able to trace where a recalled memory came from. Useful for local-first agent setups that prefer SQLite over a hosted memory service.

Requirements

Runs as a local process via uvx from the PyPI package seam-runtime. Optionally set SEAM_DB_PATH to choose the SQLite memory database location; it defaults to a per-user data directory. Optionally set SEAM_PGVECTOR_DSN to use Postgres/pgvector instead of the built-in SQLite vector store.