Keep your agent stack
Built on the open AG-UI protocol, from the team that created it.
Turn the work your users do into the skills your agents need. Learn from real workflows. Evaluate on real tasks.
Right now, your users or employees are doing by hand the exact tasks you're trying to teach your agent.
Today we're launching CLiF: Continual Learning from full-stack trajectories.
SkillResolve a duplicate charge
But your users already know how. What if they could teach your agents?
AG-UI carries what your agent does. Continual Learning adds what your users do, in the same open stream, from the team that created it.
Supported by the major agent stacks, so you keep the one you have.
Agent events and user actions land in one full-stack trajectory.
Trajectories stay in an open format you can export to your own tools.
Agent Trajectory
Product Trajectory
Agent called the billing toolAgent repliedOpened ticket #4821Issued an account creditFull-Stack Trajectory
Full-Stack Trajectory
From production traffic, or from flows your PMs record.
Works with any agent or MCP server
Real tasks, from real usage
Fully hackable, in the tools you use
A custom model tuned on your trajectories
Built on the open AG-UI protocol, from the team that created it.
Capture the paths your users take to get the job done.
Self-host CopilotKit Intelligence in your own VPC or on-premises.
Export trajectories and production-based evals to your own tools.
Product trajectories are in early access. Skill creation and Intelligence self-hosting are available today.