A Real-Time Data Layer for AI Agents
For teams building AI SDRs, research agents, or any autonomous workflow that calls a data API programmatically.
| Endpoint | Returns |
|---|---|
GET /v1/person | A full enriched profile, callable directly from an agent tool call |
GET /v1/company | A full enriched company record, callable directly from an agent tool call |
POST /v1/people/search | Structured people search, for an agent building its own candidate or prospect list |
POST /v1/people/bulk-enrich | Batch enrichment, for an agent processing a list rather than one record at a time |
POST /v1/monitors | A webhook an agent (or the system it feeds) can react to asynchronously |
An AI agent calling a data API doesn't behave like a human using a dashboard. A person doing manual prospecting makes a roughly steady, bounded number of lookups a day. An agent given a task like "research this list of 40 companies" can burn through hundreds of calls in a few minutes, then sit idle for hours, then spike again the moment a new task queues up.
That traffic shape is exactly what breaks credit-based and per-seat pricing models, which are built around predictable, human-paced usage. A credit pool sized for a person's daily habits gets exhausted in minutes by an agent working through a batch task, and a per-seat price doesn't map cleanly onto a process that isn't a person logging in at all.
Why flat-rate fits agent traffic specifically
Kooperativa's pricing has no credit pool to exhaust and no seat count to reason about. The only constraint is a shared rate limit of 500 requests per minute per workspace, sized for infrastructure stability, not as a way to meter cost by usage. An agent that needs to enrich 500 companies in one task and then nothing for the rest of the day costs the same as one making a steady trickle of calls, because the license price doesn't change with volume either way.
That also removes a failure mode specific to agent workflows: a credit-metered API that runs out mid-task forces an agent to either stop, degrade, or wait for a reset, none of which are graceful when the agent is operating autonomously without a human watching to notice and top up credits.
Schema stability matters more for a machine caller than a human one
A human using a UI can tolerate a slightly inconsistent response shape, they read around it. An agent parsing a response programmatically cannot. Every Kooperativa endpoint returns a stable, documented JSON shape with explicit null for fields a source profile never filled in, rather than omitting the key, so an agent's parsing logic doesn't need defensive code for a field that's sometimes present and sometimes entirely missing from the object.
Bulk endpoints follow the same principle: /v1/people/bulk-enrich separates matched, not_found, and failed counts explicitly in the response, rather than a flat matched/unmatched boolean, so an agent deciding whether to retry a specific identifier has the information to make that decision without guessing.
{
"tool": "enrich_person",
"input": { "linkedin_url": "https://www.linkedin.com/in/satyanadella" },
"endpoint": "GET https://kooperativa.io/api/v1/person",
"note": "No credit check before the call, no pool to exhaust mid-task."
}Get started
Try Kooperativa
One API key. Every endpoint on this page, and every other one, included on the same flat license with unlimited requests.