> ## Documentation Index > Fetch the complete documentation index at: https://aisa.one/docs/llms.txt > Use this file to discover all available pages before exploring further. # Exa Agent Runs > Submit a research task to the Exa Agent as an asynchronous job — submit, then poll the run for status and results. Submit a natural-language research `query` and the Exa Agent runs it in the background, returning a structured result. Because deep research takes time, Agent Runs is an **asynchronous job**: you submit the run and then poll it for status and results. **Lifecycle** 1. **Submit** — `POST /apis/v1/exa/agent/runs` with the research body and a required `Idempotency-Key` header. On success you get back `202 Accepted`, a `Location` header pointing at the job resource (`/apis/v1/exa/agent/runs/{jobId}`), and a job object whose `status` starts as `queued`. Re-submitting with the same key and the same request body returns the original run; the same key with a different body returns `409 idempotency_conflict`. 2. **Poll** — `GET /apis/v1/exa/agent/runs/{jobId}` and repeat until `status` is terminal: `completed`, `failed`, or `cancelled`. When `completed`, the research result is in `output`. The provider supports polling only — there is no webhook or SSE callback. 3. **List** — `GET /apis/v1/exa/agent/runs` (optional `status` filter, paginated) to page through your runs. 4. **Cancel** — `POST /apis/v1/exa/agent/runs/{jobId}/cancel` to request cancellation; cancelled runs are not billed. Supply an `outputSchema` to validate the structured output, `effort` to trade depth for latency, `previousRunId` to continue a prior run, and `input`/`dataSources` for row-processing and Exa Connect sources. Billed at a flat **\$0.10** per run. For a single-shot cited answer instead of a multi-step run, use [`post_exa-answer`](/docs/api-reference/search/post_exa-answer). ## OpenAPI ````yaml openapi/exa.json POST /exa/agent/runs openapi: 3.0.0 info: title: Exa API version: 1.0.0 description: >- Unified API documentation for Exa neural semantic search endpoints exposed through the AIsa gateway: Search, Contents, Answer, and the asynchronous research Agent Runs. Search, Contents, and Answer are synchronous and billed at a flat $0.08 per successful request. Agent Runs is asynchronous: the caller submits a run and then polls the returned job resource for status and results, billed at a flat $0.10 per run. Upstream 4xx responses (for example 401/403/404) are not billed. The `costDollars` field that Exa returns is the upstream cost breakdown and is informational only — customer billing follows the flat AIsa prices above. servers: - url: https://api.aisa.one/apis/v1 security: - bearerAuth: [] paths: /exa/agent/runs: post: tags: - https://docs.exa.ai/reference/create-a-research-task summary: Submit an asynchronous research Agent run. description: >- Asynchronous. Submit a natural-language research task; the Agent runs it in the background and returns a structured result. Billed at a flat $0.10 per run. Lifecycle: 1. Submit — POST /exa/agent/runs with the research body and a required `Idempotency-Key` header. On success the API returns HTTP 202 Accepted, a `Location` response header pointing at the job resource (`/apis/v1/exa/agent/runs/{jobId}`), and an integration_async_job body with `status` = queued. Re-submitting with the same key and request fingerprint returns the original run; the same key with a different fingerprint returns 409 idempotency_conflict. 2. Poll — GET /apis/v1/exa/agent/runs/{jobId} to retrieve the run. Repeat until `status` is terminal (completed, failed, or cancelled). When completed, the research result is in `output`. The provider supports polling only (no webhook or SSE callback). 3. List — GET /apis/v1/exa/agent/runs (optional status filter, paginated) to page through your runs on this endpoint. 4. Cancel — POST /apis/v1/exa/agent/runs/{jobId}/cancel to request cancellation. Cancelled runs are not billed. parameters: - in: header name: Idempotency-Key required: true schema: type: string maxLength: 191 description: >- Unique key (1 to 191 characters) that makes the submit idempotent. Re-submitting with the same key and request fingerprint returns the original run. requestBody: required: true content: application/json: schema: type: object properties: query: type: string description: The natural-language research query. example: >- Summarize the main changes in RAG evaluation methods over the past year as a bullet list. outputSchema: type: object description: JSON Schema used to validate the structured output. input: type: object description: 'Row-processing input: rows to process and exclusions.' properties: data: type: array items: {} description: Input rows for the run. exclusion: description: Items to exclude from processing. effort: type: string description: Compute/depth tier for the run. previousRunId: type: string description: Continue from a previously completed run. dataSources: type: array items: {} description: >- Third-party data sources (Exa Connect) the Agent is granted access to. required: - query responses: '202': description: >- Research run accepted. The Location header points at the job resource; poll it until terminal. headers: Location: schema: type: string description: >- Path of the created job resource, e.g. /apis/v1/exa/agent/runs/{jobId}. content: application/json: schema: $ref: '#/components/schemas/AsyncJob' '409': description: >- idempotency_conflict — the same Idempotency-Key was reused with a different request body. components: schemas: AsyncJob: type: object description: >- An asynchronous integration job. Returned by the submit call (HTTP 202) and by the poll/detail call. Poll the job by its id until status is a terminal value (completed, failed, or cancelled). properties: id: type: string description: Unique AIsa job identifier. Use it to poll, list, or cancel the job. example: iaj_01HZY8Q2M4K7N9V3T6W1X0B2C3 object: type: string description: Always "integration_async_job". example: integration_async_job endpoint: type: string description: The submit endpoint this job belongs to. example: /apis/v1/exa/agent/runs status: type: string enum: - queued - running - completed - failed - cancelled description: >- Customer-facing lifecycle status. queued and running are non-terminal; completed, failed, and cancelled are terminal. createdAt: type: string format: date-time description: When the job was accepted. completedAt: type: string format: date-time nullable: true description: >- When the job reached a terminal status. Null while the job is still queued or running. pricing: type: object properties: currency: type: string example: USD authorizedMicrosUSD: type: integer description: >- Amount authorized (held) when the job was admitted, in micro-USD. Agent Runs settle at the flat $0.10 per run. finalMicrosUSD: type: integer nullable: true description: >- Final settled cost in micro-USD once the job is terminal. Null until settlement. billingMode: type: string description: Billing mode for the job. example: flat_per_call output: description: >- Job result payload. Present only once status is completed. For Agent Runs this is the structured research result. nullable: true outputExpired: type: boolean description: >- True when the result has been retained past its retention window and is no longer retrievable. error: type: object nullable: true description: Present when status is failed. Null otherwise. properties: code: type: string description: Machine-readable error code. message: type: string description: Human-readable error message. securitySchemes: bearerAuth: type: http scheme: bearer ````