Persistent memory for AI agents is addressed by Stash, a self-hosted, single-binary solution that combats AI amnesia by enabling agents to remember, recall, and learn across sessions. It transforms raw observations into structured knowledge, benefiting developers building stateful AI applications by eliminating repetitive context-setting.

How It Works

Stash functions as a cognitive layer, processing agent interactions through an 8-stage consolidation pipeline. This pipeline incrementally converts raw episodes into structured knowledge, including facts, relationships, causal links, and confidence decay, leveraging Postgres with pgvector for storage. An included MCP-compatible server ensures efficient data processing and knowledge base consolidation.

Quick Start & Requirements

Setup involves cloning the repository, copying and editing the .env.example file with API keys and model configurations, and running docker compose up. This command provisions Postgres with pgvector, migrations, and the MCP server. Prerequisites include Docker and a compatible MCP-compatible agent.

Highlighted Details

• Provides persistent memory, enabling AI agents to retain context across conversations.
• Features an 8-stage consolidation pipeline for turning observations into structured knowledge (facts, relationships, patterns, wisdom).
• MCP-compatible, supporting agents like Claude Desktop, Cursor, Ollama, and OpenAI Agents.
• Self-hosted, single-binary deployment with no cloud dependency.

Maintenance & Community

No specific details regarding contributors, sponsorships, or community channels (like Discord/Slack) were found in the provided README.

Licensing & Compatibility

Licensed under Apache 2.0. This license is permissive and generally compatible with commercial use and closed-source applications.

Limitations & Caveats

The provided README does not explicitly detail limitations, unsupported platforms, or known bugs. The project appears to be self-contained and focused on its core memory persistence functionality.