Route prompt
Route a user prompt to the best candidate agent in the current project.
Authentication: personal access token. The spec defines an api_key security
scheme — send an Authorization: ApiKey <your-personal-access-token> header
(value format: ApiKey <your key>). A session_cookie scheme (connect.sid)
exists for browser sessions. Operations in this spec declare security: [], an
artifact of the upstream spec generator — authentication is still enforced by the
server.
Request body
{
"projectUuid": "string",
"prompt": "string"
}Responses
200 — Success
{
"results": {
"nextAction": "create_thread",
"decision": {
"candidates": [
{
"description": "string",
"name": "string",
"agentUuid": "string"
}
],
"reasoning": "string",
"confidence": "high",
"suggestedAgentUuid": "string",
"decisionUuid": "string"
}
},
"status": "ok"
}default — Error
{
"error": {
"data": "string",
"message": "string",
"name": "string",
"statusCode": 0
},
"status": "error"
}Upsert AI router instruction
Write a new routing-instruction version for a project. Versions are append-only; the latest one is the active instruction.
Commit router decision
Commit a router decision once the user has resolved it (router auto-routed and the thread is created, or the user picked an agent from the picker). Decisions left uncommitted indicate the user abandoned the flow.