Docs
EN/中

Guide
  • Quickstart
Integrations
  • Agent Skills
  • CLI
  • MCP
  • SDK
API reference
  • Search
  • Contents
  • Research
Concepts
  • Authentication
  • Errors
  • Rate limits
Download OpenAPI

POST /v1/research

Research

Searches, reads and answers with cited sources. fast answers within the request, deep runs in the background, and auto chooses for you.

Try it →Markdown

Request body

query
string
Required.What to search for or research. Non-empty, up to 2,000 characters.
mode
string
fast for a quick answer, deep for in-depth research, auto to choose by query.Default: auto

Response

id
string
Run id, for getting, cancelling or deleting the Run.
status
string
queued, running, completed, failed or cancelled.
created_at
string
When the Run was created.
started_at
string | null
When the Run started; null before then.
completed_at
string | null
When the Run ended; null until then.
mode
string
The mode the Run used: fast or deep.
message
string
A polling hint while the Run is in progress.
requested_mode
string
The mode you requested.
error.code
string
Why the Run failed. Present only when status is failed.
error.retryable
boolean
Whether running the same query again may succeed.
result.answer
string
The answer, once the Run is completed.
result.sources
object[]
Cited sources, each with title and url.

Runs

A fast request waits up to 20 seconds and returns 200 with the finished Run. deep, or a Run still going when the wait ends, returns 202 with its id: get it until status is completed, failed or cancelled. Send Prefer: respond-async to return at once, or Prefer: wait=N to wait at most N seconds.

GetGET /v1/research/{id}
The Run as it is now. While it is queued or running, get it again after Retry-After.
CancelPOST /v1/research/{id}/cancel
200 once cancelled, 202 while cancellation finishes. A Run that has already ended returns 409 run_not_cancellable.
DeleteDELETE /v1/research/{id}
204 once deleted. 202 while a running Run is stopped and deleted; it then returns 404.

Use the API key that created a Run for every operation on it.

See also: Authentication · Errors · Rate limits

Request
curl "https://api.arex-research.com/v1/research" \
  -H "Authorization: Bearer $AREX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"Compare current approaches to AI agent evaluation","mode":"auto"}'
Response
{
  "id": "11111111-1111-4111-8111-111111111111",
  "status": "completed",
  "mode": "fast",
  "requested_mode": "auto",
  "created_at": "2026-09-03T08:00:00Z",
  "started_at": "2026-09-03T08:00:01Z",
  "completed_at": "2026-09-03T08:00:12Z",
  "result": {
    "answer": "AI agent evaluation combines task-success benchmarks with checks of tool use, reliability, and safety.",
    "sources": [
      {
        "title": "AgentBench: Evaluating LLMs as Agents",
        "url": "https://arxiv.org/abs/2308.03688"
      }
    ]
  }
}

Example response. Actual values vary.

Previous pageContentsNext pageAuthentication