Recall Loom

RecallLoom is an open source continuity layer that lets AI coding agents pick up a project where the last session stopped. Instead of storing chat history, it keeps verified project state, key decisions and the next step as plain files in a folder inside the repository, so a new session, model or tool does not need the project explained again. The saved record is a small set of files: a context brief covering goals and boundaries, a rolling summary of current progress, and a state file that can be validated and tracked in version control. When a user tells an agent to continue the project, the agent reads these files and checks them against the Git history, the working tree and relevant tests, so a stale summary does not override the real code. Writes are explicit, with a suggested record and plan shown before anything is saved, and the vendor states it does not edit business code by default. It is installed as a skill with a single npx command and requires Python 3.10 or later. The vendor lists Codex, Claude Code, Gemini CLI and GitHub Copilot as agents that can share the same project memory. It is local first, needs no cloud service and is not a vector database or retrieval system. The site is written mainly in Chinese. The site publishes no pricing, and the project is presented as free open source software.