Execute Agent
AI-generatedSummary
Send a JSON-compatible input to a selected Watsonx Orchestrate agent through its chat completions endpoint. Optionally use a conversation context key to reuse the thread returned by an earlier execution in the same node process.
Inputs
- Selection Mode (required) — Choose an agent from the discovery list or enter its ID manually. If list selection does not resolve an ID, the node also checks the manual ID field.
- Agent / Agent ID (required) — Select a discovered Watsonx Orchestrate agent, or provide its ID when using Manual mode.
- Input (required) — A JSON-compatible value. The node sends it as a user message: strings are sent as text, numbers and booleans as strings, and arrays, objects, or null as JSON text (null becomes an empty message).
- Timeout (ms) — Per-item request timeout, from 1000 to 120000 ms; defaults to 30000 ms.
- Conversation Context Key — A stable key, such as a session or customer ID. The node uses it to look up a thread ID saved from a previous successful execution and sends that thread ID with the request. If the response contains a thread ID, it saves it under this key.
- Output Mode — Choose Chat (default), Concise, or Full. Chat returns response text and thread/request context; Concise returns the response and metadata; Full also includes the raw provider response.
Output shape
One output item per input item, containing a JSON object.
In Chat mode, the object contains response, text, threadId, sessionId, sentThreadId, and requestId. Concise and Full return response, raw, and metadata; raw is populated only in Full mode. The response is extracted from the first choice's message content or text, then from output, result, or response fields, falling back to the raw response. If Continue On Fail is enabled, failed items contain structured error details.