Living Memory

Most AI memory tools solve recall. Living Memory also solves handoff. WHAT IT IS Living Memory is a shared memory world your AI agents connect to over MCP - Claude, ChatGPT, Codex, Cursor, a CLI, or your own bot. Tell one agent something today; ask a different agent tomorrow and it already knows. The memory belongs to the project, not to a single vendor, device, or chat session. WHAT MAKES IT DIFFERENT: AGENTS CAN LEAVE WORK FOR EACH OTHER Memory holds what should last — a decision, a promise, a deadline, a dead end you already ruled out. Handoff holds work in flight. A handoff note carries up to 64KB of raw text, passed through verbatim: a spec, a half-finished migration, an exact error message. The next agent reads the exact bytes, not a summary. Notes expire on their own after 24–72 hours, so temporary work state never pollutes long-term memory. That is the part no other memory tool does. > Recall means one agent remembers. > Handoff means one agent can finish what another one started. WHAT YOUR AGENT GETS - memory_add — store a durable fact, decision, or correction - memory_search — semantic recall across sessions and models (MMR retrieval, not keyword matching) - memory_state — see what the world currently holds - memory_forget — drop a fact that has been superseded - handoff_post / handoff_read / handoff_list — pass exact working state between agents - world_list — one shared context per project, not per person and not per tool - client_mint / client_revoke — invite an agent that cannot sign in with OAuth (a CLI, a cron job, a custom bot), and cut its access off when you are done SETUP One MCP server URL. No SDK and no code — it works anywhere MCP works. WHO IT IS FOR Anyone running more than one AI agent and tired of re-explaining the same project every morning. Solo builders switching between Claude and ChatGPT. Teams where the human is the only one who was present for the whole project. Agent workflows that span machines, sessions, and vendors. Free to start .