CopilotKit Releases CLiF for Continual Learning From Agent and User Trajectories

CopilotKit is the company behind the open-source CopilotKit framework and the AG-UI protocol, MarkTechPost reported. Its open-source project has more than 38,000 stars on GitHub. CLiF is part of CopilotKit Intelligence and is built around the full-stack trajectory as its central primitive.

MarkTechPost reported that teams shipping AI agents at scale keep running into the same pattern. Even agents built on top-tier models such as Fable and Astra struggle to carry out complex, end-to-end tasks inside a specific product consistently enough to deploy. Teams write evals for the workflows that matter most, often hundreds or thousands of hand-written cases, then improve the agent through prompt edits, harness fixes, and special cases. Meanwhile, users carry out those same tasks every day and know which actions to take and in what order.

The platform is designed to address a gap in most production agent stacks. Agent observability today centers on the agent, capturing prompts, tool calls, and model outputs. That shows where an agent went wrong but leaves the correct path out of view. Product analytics covers clicks, funnels, and session replays, but that data is made for a person reading a dashboard, not for teaching an agent a task or testing it. In most stacks, the two sides never end up on the same timeline, according to MarkTechPost.

CLiF captures both sides through AG-UI, the open protocol CopilotKit created to connect agents to user-facing applications. Because AG-UI sits between the agent and the user, it can record the agent trajectory, including every run, tool call, and state change, and what each returned. It also records the product trajectory, including what was on the user’s screen, the buttons they clicked, and the server requests they triggered. If the user talks to the agent during the task, that conversation is placed on the same timeline in order. The result is a full-stack trajectory. AG-UI already streams agent events to the frontend, so capturing user actions over the same protocol puts both halves in one stream without a separate analytics pipeline to stitch them together later, MarkTechPost reported.

Each user session becomes a single trajectory that a team can open in CopilotKit Intelligence and replay step by step, with clicks, server requests, and agent messages in the order they happened.

MarkTechPost reported that CopilotKit describes the learning loop in four stages. In the first stage, capture, trajectories stream into CopilotKit Intelligence and are assigned to learning containers. These containers scope learning to one kind of work, such as billing support or customer onboarding. The runtime picks a container for each session through the getTrajectoryLearningContainerIds callback, and sessions can later be moved in or out of a container in the Intelligence UI. Each container learns only from its own sessions.

In the second stage, find patterns, analysis runs on a schedule after a container passes a set threshold of new sessions, and teams can also trigger it manually. Each run works from a frozen snapshot of the container’s sessions and surfaces insights such as recurring requests, repeated corrections, and friction that users quietly work around. CopilotKit sees that friction as the strongest signal, according to MarkTechPost.

The broader four-stage loop, built on CopilotKit’s Automatic Learning, turns full-stack trajectories into skills, production-based evals, fine-tuning datasets, and raw exports, with a human reviewing every skill, MarkTechPost reported. A worked example shows how agents can learn the way team leads actually close out a sprint.

CLiF targets several pains that slow agent teams, including evals, context bloat, cost, fragmented stacks, and shadow AI usage, according to MarkTechPost. The platform runs fully self-hosted.