Research answers with cited sources.
Give Superagent a research objective and the fields you need back. It searches and reads public sources, follows up on missing information and returns structured findings with source URLs and supporting quotations. Your application starts a task and collects the result later.
curl https://api.cloud.gomry.com/v1/superagent \
-H "Authorization: Bearer $GOMRY_CLOUD_KEY" \
-H "Content-Type: application/json" \
-d '{
"objective": "Find the published cancellation terms.",
"seedUrls": ["https://example.com/supplier/terms"],
"outputSchema": {"type": "object", "properties": {
"cancellationTerms": {"type": "string"}
}},
"tier": "standard"
}'Follow a question beyond the first search.
When a supplier’s cancellation terms are spread across several pages, Superagent can use what it has read to identify unanswered parts of the question and look for supporting information in further public sources. The research stays within the task’s work limits.
A procurement app could request the cancellation terms in a cancellationTerms field. In the illustrative result below, the app displays the finding beside its supporting passage and source. With a real result, a reviewer could open that source to check the answer before making a purchasing decision.
- Finding
- Cancellations at least 14 days before the booking receive a full refund.
- Supporting passage (synthetic)
- “Cancel at least 14 days before your booking to receive a full refund.”
- Source (illustrative URL)
- https://example.com/supplier/terms
When sources disagree, the result keeps both versions and their supporting passages for review. Information without a supporting source is left out. Superagent reads public pages and does not sign into private accounts.
Start the task now; collect the research when it finishes.
POST /v1/superagent returns a task ID, not the completed research. Work may continue for minutes or hours. Poll GET /v1/superagent/{taskId} at the response’s Retry-After interval, or provide a webhook URL for notification when the task reaches a terminal state.
{
"objective": "Find the published cancellation conditions for this supplier.",
"seedUrls": [
"https://example.com/supplier/terms"
],
"outputSchema": {
"type": "object",
"properties": {
"cancellationTerms": {
"type": "string"
}
}
},
"tier": "standard"
}Use an Idempotency-Key when your application might retry task creation, so a retry does not start another research task. Polling and cancellation are not billed; cancellation stops further work rather than refunding work already consumed.
Choose a work limit, not a fixed-price package.
| Tier | Source ceiling | Reasoning-round ceiling |
|---|---|---|
| Lite | 3 sources | 0 rounds |
| Standard | 15 sources | 2 rounds |
| Deep | 100 sources | 6 rounds |
A task can finish before reaching those limits. Your own budget can narrow the tier’s cost, source, time and round ceilings. For a single known page that needs extraction during one request, use Scrape instead.
Task, source and round are separate charges.
| Unit | Rate |
|---|---|
| Task | $0.15 |
| Source read | $0.02 |
| Reasoning round | $0.05 |
For illustration, a task that reads five sources and uses one reasoning round costs $0.30: $0.15 for the task + (5 × $0.02) for sources + $0.05 for the round. The total depends on the work actually used.
Polling an existing task and cancelling it add no fee. Usage accompanies the task so your application can track the work and spend together.
Build with Superagent.
Enable Superagent for the project and grant task creation and reading permissions to the key your worker will use. Send the key as a Bearer token. Configure polling or a public webhook in your application before starting background research.