Actions8
- Agent Actions
- Answer Actions
- Content Actions
- Search Actions
Agent → Create Run
AI-generatedSummary
Start an Exa Agent run to research, build a list, or enrich data from a natural-language task. Choose text output or validated JSON, optionally provide input records, data sources, effort, metadata, and a system prompt, then either wait for completion or return the run immediately for later retrieval.
Inputs
- Query (required) — Natural-language task for the agent.
- Output Format — Choose a cited natural-language report or structured JSON matching the supplied Output Schema.
- Output Schema — JSON Schema used when structured output is selected.
- Wait for Completion — Defaults to waiting. Turn off to return the newly created run immediately and retrieve it later with Get Run.
- Wait Mode — When waiting, choose a server-sent events stream or periodic status polling.
- Poll Interval (Seconds) — Delay between status checks when using Poll mode; defaults to 4 seconds.
- Include Events — In Stream mode, attach collected streamed events alongside the finished run; defaults to false.
- Timeout (Seconds) — Maximum wait time when waiting for completion; defaults to 600 seconds.
- Options — Optional continuation run ID, up to five Exa Connect data sources, effort level, JSON array of input records, metadata JSON, and system prompt. Input Data must be a JSON array; metadata and structured schema must be valid JSON.
Output shape
One n8n item containing the run object; in Stream mode with Include Events enabled, the object also contains an events field.
With waiting disabled, returns the run response immediately. With waiting enabled, returns the terminal run after completion, failure, or cancellation. Polling or streaming can fail if the configured timeout is reached before a terminal state.