AI Research & Development
Building infrastructure for
intelligent systems
We build software that lets people keep their AI agents' memory on every machine they use, encrypted so that only they can read it.
Products
Cordelia
Pre-releaseYour agent's memory on every machine you use
AI coding agents keep memory, but it lives in one folder on one machine. Cordelia keeps it in step across your devices. It is encrypted on the device that wrote it and carried by relays that cannot read it, and a project's memory follows the project wherever it is cloned. Cordelia works with Claude Code.
Thesis
Agent memory is stuck where it was written
Agents remember preferences, decisions and the reasons for them. That memory is filed by machine and by path, so a second machine, or a second clone of the same project, starts from nothing.
Memory must be encrypted end to end
Agent memory holds business decisions, personal preferences and private knowledge. It deserves the protection people expect for their messages: encrypted on the device, and readable only by the people it belongs to.
Infrastructure should hold no keys
The services that carry memory should not be able to read it. Cordelia's relays store only ciphertext. Anyone can run one, and none of them, ours included, can read what it carries.
Say only what is built
We keep the product small and describe only what it does. Its threat model sets out what it protects against and what it does not. The code is open source, under AGPL-3.0.
About
Seed Drill Labs is a research and development studio in the UK. We build infrastructure for AI agents, starting with Cordelia, which keeps an agent's memory in step across the machines its owner uses, encrypted end to end.
Our work draws on distributed systems, information theory and cryptography. We publish our research, and our code is open source under AGPL-3.0.