Recall™

Recall™ is an open reference architecture for durable memory, audit, and continuity in AI coding agents. It is designed around the problem that modern agents are stateless: each conversation starts cold, and memory, audit, and long-term context depend on what the host platform keeps or a chat transcript preserves. The site describes it as a way to turn an agent into something that reads a project log before acting, rather than relying on its own short session history.

Its documented components are memory, local compute, vault, guardrails, brain backup, and continuity. The architecture uses hot, warm, and cold context tiers and a networked-brain pattern. The brain is described as a vector database that can scale to tens of thousands of chunks of knowledge, and the site says the log lives on disk, in a vector brain, and in immutable cloud storage. It also mentions a cold-start protocol, overnight backup, and a small set of conventions that work with about 600 lines of glue.

The reference implementation is shown with VS Code, GitHub Copilot Chat, an object-storage backend, and a local open-weights model running on a consumer GPU. The site also says it is not bound to those components and can be swapped freely. A quickstart is provided with three commands, along with links to the whitepaper, GitHub, and comparison material that references Cursor, Cline, Aider, Continue.dev, LangChain, and AutoGen.

Recall™ is free to read, free to use, and free to adapt. The whitepaper is shared freely, and the reference implementation is permissively licensed under MIT. The site also notes that attribution is appreciated and that the trademark application is pending.