Plenty of extraction lives inside its own UI — a place people log into, upload files, and copy results out of. That’s fine for a standalone workflow and a wall for everything else. The real leverage in a carrier comes when extraction runs inside the systems that already do the work: the moment a submission hits the workbench, the instant an FNOL lands, extraction should just happen, and the structured result should flow straight into the next step.
The agentic API is that surface. It exposes the same engine the console uses — classification, the reasoning loop, multi-document consolidation — as a service your systems call directly.
One call, the whole pipeline
You submit a bundle of documents for an account. The API detects the line of business, classifies each document, extracts, consolidates to the output contract, and returns the record — you don’t orchestrate the steps.
POST /api/v1/agentic/extract Authorization: Bearer <api-key> Idempotency-Key: sub-88431 { "account_ref": "SUB-88431", "documents": [ { "name": "broker-email.pdf", "url": "s3://.../broker-email.pdf" }, { "name": "acord-125.pdf", "url": "s3://.../acord-125.pdf" }, { "name": "loss-run.pdf", "url": "s3://.../loss-run.pdf" } ], "callback_url": "https://your-app/hooks/ix" // optional — or poll }
{
"account_ref": "SUB-88431", "lob": "excess_casualty",
"status": "completed", "overall_confidence": 0.94,
"record": { "insured_name": "Acme Logistics, LLC", /* … contract fields */ },
"provenance": { "insured_name": { "source": "acord-125.pdf", "page": 1 } },
"conflicts": [], "flags": [ { "field": "effective_date", "reason": "low_confidence" } ]
}+ provenance + confidence] RES -->|webhook / poll| APP
Built for machine-to-machine
| Concern | How the API handles it |
|---|---|
| Auth | API keys or OAuth client-credentials, scoped per organization. |
| Safety on retries | Idempotency keys — resubmitting a bundle doesn’t double-process it. |
| Throughput | Batch submission; asynchronous processing with poll or callback. |
| Completion | A webhook fires when a run finishes — no busy-waiting required. |
| Stability | A versioned contract, so integrations don’t break under you. |
| Isolation | Per-organization scoping keeps each tenant’s data and config separate. |
The response carries everything a downstream decision needs: the record shaped to your output contract, per-field provenance and confidence, any unresolved conflicts, and the fields flagged for review — so your system can auto-post the clean ones and route the rest.
Why an API surface matters
- Extraction where the work happens. A submission triaged in your workbench or an FNOL in your claims system gets extracted in place — no swivel-chair to a separate tool.
- Straight-through, end to end. With confidence and conflicts in the response, your automation can post high-confidence records itself and queue only the exceptions.
- Use the agents independently. The same engine that powers the console is callable on its own — embed it in a pipeline, a bot, or an orchestration layer you already run.
- Integrate in days. One authenticated endpoint, a versioned contract, and a webhook — not a bespoke integration project.
The console and the API are one engine
Everything a person can do in the UI — classify, extract, consolidate, resolve, validate — is available programmatically, against the same configuration. So a document type you build in the builder is immediately callable over the API, with no separate wiring.
Read next →
See the other way work comes in: folder-native intake and round-trip, and the governance around who can call what: governance-first: Org → Business Unit → Project.