BIMLabz

MemoryLoom is one shared memory for a team of AI agents. I work with several agents on different projects. Each agent starts every session with no memory of what the others learned. I end up repeating the same facts, and work gets done twice or done against a decision that was already changed. MemoryLoom fixes that. An agent writes what it learned into one shared store. Any other connected agent can recall it later, in a new session, on a different project. The person tells one agent something once, and every connected agent can answer from it later. Nothing enters the shared store directly. Every suggestion passes a gate first. The gate checks the claim, and some claims are sent to a human instead of being approved. A suggestion with no source waits for a person. That is happening on the live site today. Four suggestions are waiting now for a person to decide. Every record keeps its proof. A recall answer shows who wrote the record, when it was recorded, and what it rests on. If the store holds no answer, it says so in plain words and shows the closest records marked as records that do not answer. MemoryLoom is live now at memoryloom.vercel.app. Its store holds real records from real work on my own projects, one each from five projects, written and approved on 4 November 2026. Agents connect to it over MCP, the same open connector my tools already use, at memoryloom.vercel.app/mcp. The connector is guarded by an owner word, so only my agents can write to my store.