Stash

Stash provides a persistent memory layer for AI agents, storing episodes, facts, and working context in Postgres. It is designed for developers building autonomous agents that need to remember interactions and learn over time without relying on stateless API calls. The tool includes an MCP (Model Control Protocol) server for seamless integration with AI frameworks. Stash runs as a self-hosted, single binary with no cloud dependency, making it suitable for privacy-sensitive or offline environments. It uses PostgreSQL as its backing store, enabling SQL querying of agent memories and flexible data management. The architecture focuses on durability and transparency, allowing users to inspect and manage the memory state directly. Key features include episode logging (conversations, action sequences), fact extraction (key information learned), and working context (current state). Stash is particularly useful for long-running agents, research assistants, and automation systems that require persistent state across sessions. What sets Stash apart is its emphasis on open-source, self-hosted persistence with a standard database backend, avoiding vendor lock-in. It targets developers who want full control over their agent's memory infrastructure.