Quickstart

This example invokes an existing workflow revision. You need an Agenta host, project ID, API key, and workflow revision ID.

export AGENTA_HOST="https://eu.cloud.agenta.ai"
export PROJECT_ID="019d952f-0000-0000-0000-000000000000"
export REVISION_ID="019d952f-0000-0000-0000-000000000002"

Run the first turn

curl -iN -X POST \
  "$AGENTA_HOST/services/agent/v0/invoke?project_id=$PROJECT_ID" \
  -H "Authorization: ApiKey $AGENTA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: text/event-stream" \
  -d "{
    \"references\": {\"workflow_revision\": {\"id\": \"$REVISION_ID\"}},
    \"data\": {\"inputs\": {\"messages\": [
      {\"role\": \"user\", \"content\": \"Inspect the project and explain how to run its tests.\"}
    ]}}
  }"

Response headers include x-ag-session-id, x-ag-trace-id, and x-ag-span-id. Each SSE data: line contains an Agenta event. A final done event reports the stop reason.

Continue the session

Send another request with the same revision and session_id:

{
  "session_id": "<x-ag-session-id from the first response>",
  "references": {
    "workflow_revision": { "id": "019d952f-0000-0000-0000-000000000002" }
  },
  "data": {
    "inputs": {
      "messages": [
        { "role": "user", "content": "Now run the smallest relevant test." }
      ]
    }
  }
}

If a turn pauses, answer its pending interaction before continuing. See Tools and interactions. To stop an active execution, call POST /api/sessions/{session_id}/cancel.

For a non-streaming result, send Accept: application/json. For Vercel AI SDK frames, also send x-ag-messages-format: vercel.