This page shows how the Conversation API turns a message into a durable run and stores the result back into the conversation.Documentation Index
Fetch the complete documentation index at: https://veryfront.com/docs/llms.txt
Use this file to discover all available pages before exploring further.
How it works
| Step | What happens |
|---|---|
| 1 | A user message is stored in the conversation |
| 2 | A run starts to produce the assistant response |
| 3 | Events record streaming output, tool calls, state, and terminal status |
| 4 | The final answer is stored back as an assistant message |
Detailed flow
What persists
| Durable object | What it stores |
|---|---|
| Conversation | Thread metadata, project scope, active branch, and message count |
| Message | User prompts, assistant answers, tool output, files, citations, and other typed parts |
| Run | Durable execution identity, owner, status, parent/root lineage, and terminal state |
| Agent run | Conversation-specific execution state, assistant message link, input anchor, and waiting checkpoint |
| Run event | Ordered execution history for streaming output, tool calls, state, errors, and finalization |
| Input request | Structured human, agent, or system input needed while a run is waiting |
Why events matter
Run events are the bridge between live execution and durable state. They let the API:- Stream output while work is active
- Replay output after a browser reconnect
- Resume work after a tool result or structured input response
- Build or repair the final assistant message
- Explain why a run completed, failed, or was cancelled