A persistent, self-managed memory for artificial minds — so that what you spoke of yesterday is still faithfully known to-day, and the long thread of an acquaintance is never severed by the throwing of a switch.
To-day's thinking engines keep but a rolling buffer of the present moment. Restart the apparatus and all is lost — the names, the deadlines, the preferences, the whole history of your acquaintance, vanished as a dream upon waking. Each conversation begins again from nothing.
No single trick, but a complete machine — each stratum tending a different faculty of remembrance, and all governed by the engine itself.
Tracks I through IV are complete and presently running. What you read below is no promise — it is in operation.
The Recallatron is a persistent memory server for AI agents and large language models (LLMs) — purpose-built to end the statelessness that afflicts every thinking engine. Whether your agent runs on Claude, ChatGPT, GPT-4, Gemini, or any other LLM, each conversation currently begins from nothing. The Recallatron ends that.
It exposes a Model Context Protocol (MCP) server, allowing any MCP-compatible AI agent or Claude Desktop installation to read from and write to its memory autonomously. Five strata work in concert: an append-only episodic memory ledger, automated session digests, cross-referenced topic threads, a versioned knowledge graph of entities and relationships, and 26 MCP tools by which the LLM agent tends its own archive.
Built for developers building AI agents with long-term memory, teams running persistent Claude or ChatGPT workflows, and anyone who needs their LLM to remember — across sessions, across weeks, across the life of a project.
The engine is built and running; the public exhibition is forthcoming. Leave your card below and you shall be the first summoned when the gates are raised.