AutoMem is an open-source long-term memory system for AI assistants, giving them memory that survives beyond a single chat session. Published on GitHub by Very Good Plugins with about 820 stars under the MIT license and updated in October 2026, it combines a graph database with a vector store to recall decisions, relationships, and context.
When you ask an assistant “Why did we choose PostgreSQL?”, a plain vector search returns similar-sounding snippets. AutoMem also records typed relationships between memories, so it can follow the connections between a decision, its rationale, and its consequences, and return the reasoning rather than just matching words. Recall happens directly through its graph and vector index with no LLM call in the middle, keeping retrieval fast and avoiding extra generation charges; embeddings can come from Voyage, OpenAI, or a local provider. It works with assistants such as Claude, Cursor, Codex, and ChatGPT.
A distinctive design choice is that one memory store can serve many assistants at once: context saved while working in a coding tool can be recalled later in a chat assistant, because memories live in AutoMem rather than inside any single product. Optional enrichment can add more structure to stored memories over time, while day-to-day recall stays a direct index lookup. For privacy-conscious users, embeddings can be produced by a local provider, keeping the whole memory stack under their own control.
AutoMem is open source under the MIT license and can be self-hosted for free. The project site at automem.ai also offers managed ways to run it, making the overall model freemium.
AutoMem is for developers and power users who are tired of re-explaining context to AI assistants, and for teams who want shared, inspectable memory infrastructure for agent workflows rather than a black-box feature.