CopilotKit OpenGenerativeUI

An interface, an interaction, and a continued conversation.

The native workflow generated an interactive counter, changed its value from zero to one, and sent the count back to the same conversation. The assistant displayed the declared follow-up response.

What was verified

All thirteen declared expectations were met in the selected workflow.

1. One initial chat submission and one widget follow-up reached the native application.
2. Browser and agent inputs matched the controlled provider records.
3. All three application streams and all provider streams completed successfully.
4. The application displayed both declared assistant responses.
5. Native planning and generated-interface tools matched the streamed activity and agent state.
6. Generated HTML, CSS and JavaScript rendered a counter whose button changed zero to one.
7. The iframe sandbox and content security policy matched the selected configuration.
8. The widget sent its count back through the intended bridge into the same conversation.
9. Independent agent-state readback matched the selected conversation and preserved the empty control thread.
10. Four native model requests stayed within the declared limit of twelve.
11. Two provider tool proposals stayed within the declared limit of four.
12. Provider output reservations totaled 8,101 bytes within the declared 262,144-byte limit.
13. Observed provider model work totaled 16,650 milliseconds within the declared 180,000-millisecond limit.

Coverage of this run

Nine established scenarios from this run’s selected inventory of nine:

• Native browser submission
• Application provider invocation
• Completed streams and visible responses
• Planning and generated-interface tool calls
• Interactive generated widget
• Selected iframe configuration
• Widget follow-up into the same conversation
• Agent state and exact readback within one server lifetime
• Declared invocation, output and time budgets

The nine-scenario denominator belongs to the selected counter and follow-up workflow. The thirteen met expectations describe its observed behavior in the recorded configuration.

This baseline uses a Factory synthetic provider and agent state within one server lifetime. Production configuration, live-provider behavior, semantic answer quality, broader authorization and multi-turn memory are further verification scopes. The iframe assessment establishes its selected configuration.

Supporting evidence

Five original browser captures from the recorded run, followed by curated summaries of its seven evidence records.

Recommended next verification

Verify the configured live-provider workflow.

Select the provider and model, accepted response-quality criteria and a bounded evaluation budget.

Extend generated-interface coverage.

Select additional widgets and interactions, then connect generated tool payloads to streamed activity, rendered state and intended bridge calls.

Extend agent integration coverage.

Select skill-file reads, explicit agent-to-MCP calls and additional subagent scenarios with their own evidence.