Open source memory layer so any AI agent can do what Claude.ai and ChatGPT do
8 hours ago
- #Cognitive Layer
- #AI Memory
- #Agent Learning
- Stash serves as a persistent cognitive layer for AI agents, enabling continuous memory across sessions without replacing the AI model.
- It organizes learned information into hierarchical namespaces (like folders), allowing automatic inclusion of sub-paths for efficient data retrieval.
- Unlike RAG, which is a smarter search engine over static documents, Stash learns from conversations, decisions, successes, and failures, synthesizing observations into a deepening knowledge graph.
- Stash is platform-agnostic, working with any AI model via MCP compatibility, unlike locked-in memory features from platforms like Claude.ai or ChatGPT.
- It addresses the limitations of AI models that forget everything between sessions, eliminating the need to repeat information and avoiding repeated failures.
- Setup is easy using Docker Compose, with no manual infrastructure or dependencies, and it supports various backends (e.g., OpenAI-compatible, OpenRouter, Ollama) without vendor lock-in.
- A background process continuously synthesizes experiences into structured knowledge, and tools cover the full cognitive stack, from basic memory functions to advanced reasoning like contradiction resolution.
- Stash is open-source (Apache 2.0 licensed), backed by PostgreSQL, and includes features like a self-namespace for agents to model their own capabilities and automated research loops for ongoing learning.