Skip to content
ServiceOps

AI data sources

Some answers do not live in articles: “what is the status of order 4512?”, “is the billing service up right now?”. Data sources let the AI query your own HTTP APIs in real time to answer them.

The model: you declare, ServiceOps executes

Section titled “The model: you declare, ServiceOps executes”

You describe an endpoint — URL, method, parameters, authentication — and when it is useful. When a request matches, the AI decides to call it, but the execution is done by ServiceOps, strictly within your declaration: your URL, your parameters, nothing else. The AI never composes raw requests, and your credentials are never shown to the model.

Create a data source

Section titled “Create a data source”
  1. Open Organization → Settings → AI → Data sources.
  2. Describe the endpoint:
    • Name and description — “Order status. Use when a customer asks about an order.” The description is how the AI knows when to use it: write it like an instruction.
    • URL with typed variables: https://api.your-company.com/orders/{orderId}
    • Method and optional headers.
    • Authentication — API key or bearer token, stored encrypted; header values never appear in logs or AI context.
    • Parameters — each variable’s type and description, so the AI fills them correctly.
  3. Test it from the form with sample values, and check the response.
  4. Save and enable.

Example

Section titled “Example”
Name         Service health
Description  Current status of a named internal service.
             Use when someone reports something "down" or "slow".
URL          https://status.your-company.com/api/service/{serviceName}
Method       GET
Auth         Bearer token (stored encrypted)
Variable     serviceName — the service's short name, e.g. "billing"

A customer writes “the billing page won’t load” → the agent calls the source with serviceName=billing → the JSON answer (say, {"status":"degraded"}) grounds the reply: “We are indeed experiencing degraded performance on billing — our team is on it.”

Security guardrails

Section titled “Security guardrails”
  • Encrypted credentials, write-only after saving.
  • Egress protection: requests can only reach the host you declared — redirects and DNS tricks cannot re-route a call to internal infrastructure.
  • Responses are data: what your API returns is treated as content to quote, never as instructions for the AI to obey.
  • Timeouts and size limits keep a slow endpoint from blocking replies.