Knowledge and learning
Persistent sources, changing information, reusable procedures, and evaluation of whether experience improves later work.
Software research & development · Kentucky, USA
More capable agents.
More dependable work.
Red Rook AI is building Keep, an experimental autonomous-work platform that brings together native execution, persistent knowledge, reusable skills, and mechanisms for controlling actions and evaluating outcomes.
Keep is under active development.
A public research preview is being prepared.
01 / The whole system
Keep is the platform.
The research spans the whole system.
Useful autonomous work takes more than a convincing response. An agent needs relevant knowledge, a way to act, clear limits on its authority, and a way to determine what actually happened. Keep brings these concerns into one experimental platform, with several research areas developing alongside the implementation.
Execution and coordination
Native repository workflows, controlled tool use, delegated authority, interacting operations, and recovery after interruption.
Control and evidence
Privacy, deployment-specific protections, outcome checks, and records that make the boundaries of a result inspectable.
02 / Work & evidence
Begin with useful work.
Judge it by the outcome.
Keep's recorded development work includes repository changes checked against defined outcomes, coding tasks using retained and corrected information, and controlled recovery experiments. Each result has a specific configuration and scope; broader capability and comparative performance remain subjects of evaluation.
See current development and evidence03 / The founder
Founder-led research
and development
Red Rook AI is led by Lisa Kaelin, whose experience includes hands-on AI evaluation, data-quality work, and human-data team leadership. She directs Keep's AI-assisted development with an emphasis on measurable outcomes and accurately stated limitations.
Meet LisaAn invitation
Research worth testing.
Systems worth building.
We welcome conversations about research support, compute access, independent evaluation, and technically meaningful collaboration.
Discuss the work