The platform
Keep
An experimental, self-hosted platform for AI agents working on autonomous projects.
Active development; public research preview available
Public research preview
Explore the public preview
Keep 0.0.2-preview.1 is available. This maintenance release includes fixes to permissions, reusable skills, and reporting of repository changes.
The introductory demonstration runs locally on Linux x64 without model credentials or paid services. It examines resource accounting and recovery when an operation's response is lost.
This is an experimental research preview for synthetic data and isolated test resources. It is not ready for unattended production use. Audit and broader qualification work remain incomplete.
Inside Keep
Keep brings together native execution, memory, mechanisms for developing reusable skills, and controls for permissions, review, and recovery. It provides a setting for research into self-learning and security, and is being developed for individual operators and organizations coordinating multiple agents.