Workflows

Beyond freeform chat, the agent-service exposes task-shaped runs: a generic agent/runs entry point and named graph/* workflows. All stream SSE except multi-persona analysis, which returns JSON.

Generic agent run

POST /api/agent/runs runs any registered task and streams it.

{
  "taskType": "analyze",
  "payload": { "resumeData": { "basics": { "name": "Ada" } } },
  "config": { "model": "gpt-4o-mini" }
}

Streams the same event types as chat (run_started → message_chunk → run_completed → done). The runId is persisted.

Inspect a run afterward

curl http://localhost:3112/api/agent/runs/clx123/events | jq

Graph workflows

Named workflows under /api/graph/*. Each takes a JSON body (commonly { resumeData, config, language }) and streams SSE.

curl -N -X POST http://localhost:3112/api/graph/analyze \
  -H "Content-Type: application/json" \
  -d '{ "resumeData": { "basics": { "name": "Ada" }, "work": [] }, "language": "en" }'

curl -N -X POST http://localhost:3112/api/graph/rewrite \
  -H "Content-Type: application/json" \
  -d '{ "resumeData": { "work": [{ "summary": "did stuff" }] }, "language": "en" }'

curl -N -X POST http://localhost:3112/api/graph/research \
  -H "Content-Type: application/json" \
  -d '{ "role": "Senior Frontend Engineer", "config": {} }'

Multi-persona analysis (JSON)

POST /api/graph/analyze-resume-multi runs several reviewer personas and returns a single JSON result (not a stream).

{
  "resumeData": { "basics": { "name": "Ada" }, "work": [] },
  "config": {},
  "language": "en"
}

It accepts resumeData (or resume), an optional config, and language (defaults to en). Because it’s not SSE, just await res.json() and read data.