Stash Ai Memory

Stash gives your AI agents a Postgres-backed memory layer that persists across sessions. It stores raw episodes, then runs an 8-stage consolidation pipeline that extracts facts, relationships, causal patterns, and goals from observations. Works with Claude Desktop, Cursor, Windsurf, and anything MCP-compatible. The Docker setup is straightforward: clone, add your LLM API key to .env, and compose up. You get semantic search over memories, working context storage, and automatic confidence decay on stale facts. If you're building agents that need to remember user preferences or learn from past interactions instead of starting fresh every time, this is a solid self-hosted option that doesn't lock you into a vendor.