CopilotKit launches CLiF for agent learning from user actions
CopilotKit released CLiF, a platform for continual learning from full-stack agent trajectories, in early access. The tool captures both agent actions and user interactions in a product, combining them into a single timeline via AG-UI, an open protocol the company developed. This data is then processed into skills, production-based evals, fine-tuning datasets, and raw exports, all versioned and…
Key points
- CLiF combines agent traces and user actions into a single timeline via AG-UI, capturing full-stack workflows for learning
- Outputs include skills, production-based evals, fine-tuning datasets, and raw trajectory exports—all human-reviewed and versioned
- Self-hostable on Kubernetes/AWS or cloud-managed, with support for frameworks like LangChain, Mastra, and Google ADK
The core idea is that most agent failures stem from missing the actual workflows users follow—often undocumented shortcuts or fixes—while traditional tracing only logs agent mistakes. CLiF addresses this by analyzing recurring patterns in user behavior (e.g., post-sprint issue adjustments) and turning them into reusable skills. It supports self-hosting on Kubernetes, AWS, or on-premises, with compatibility across frameworks like LangChain, Mastra, and Google ADK. Early adopters can also generate eval candidates and fine-tuning data from real usage, reducing reliance on manual test cases.
CopilotKit, which maintains the open-source CopilotKit framework (38,000+ GitHub stars), positions CLiF as a solution to common agent deployment challenges: eval bottlenecks, context bloat, cost inefficiency, and fragmented tooling. The platform is designed for teams shipping agents at scale, whether internal or via MCP servers for external agents like Claude.
Model page: Astra →
CopilotKit Releases CLiF: A Continual Learning Platform That Turns Agent Traces and User Actions into Skills, Evals, and Fine-Tuning Data
MarkTechPost · 15 October 2026
Loading the full article…
This text was published by MarkTechPost and written by Asif Razzaq. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Voxel51 launches 3.0 platform for physical AI data workflows · 1 src
- AWS outlines Trust in Depth for safer AI agent autonomy · 1 src
- Microsoft readies Windows for Nvidia RTX Spark laptops with agent start-up · 1 src
- Meta's Muse AI permits explicit adult conversations, unlike other Meta apps · 3 src
- Proof introduces x401 protocol to verify AI agent transactions · 1 src
Comments
via GitHub Discussions