How agents should use AI service status
Use provider reports as one input to agent recovery. Preserve request evidence, check report scope and freshness, and avoid unsupported outage or failover decisions.
Quick answer
An agent should treat a status report as supporting evidence for a failed operation. It should retain the original error, check the correct product and timestamps, then follow an application-approved recovery policy. A normal report is not permission to retry indefinitely, and an incident is not proof of every request’s cause.
Capture the request evidence first
Keep the UTC time, endpoint, model, route or provider if known, HTTP status, error code, request ID and elapsed time. For streaming, record whether output completed or ended with an error. An application can then compare repeated failures without losing the original evidence.
Store credentials and private prompt content only under your application’s own controls. IsAnythingDown does not accept API keys or ingest an agent’s request logs. The fields below are a suggested local record, not an implemented IsAnythingDown upload format.
| Field | Why keep it |
|---|---|
| occurred_at + endpoint | Place the failure in time and identify the operation |
| model + route/provider | Distinguish direct access from a gateway path |
| http_status + error_code | Separate access, limits and transient failures |
| request_id + duration | Support provider investigation and timeout checks |
| stream_completed | Avoid treating headers or partial output as completion |
Read the report’s scope and age
Use the source link and collected time on the service page. Confirm that its components describe the operation you used. ChatGPT and OpenAI API are separate; Gemini app and Gemini API are separate; a gateway and its underlying model provider are separate.
Keep stale, missing and unavailable data explicit. IsAnythingDown retains the previous valid provider report after a collection failure, so historical content can still appear. Read the page’s stale warning before citing that content as current evidence.
Apply your own recovery policy
A useful policy can stop on authentication or billing failures, pace transient retries and send unresolved failures for review. These are application decisions. Status alone cannot select a safe fallback for a task with requirements for tool support, output format, cost or data handling.
If your application permits a different model or provider, verify those requirements and route constraints before switching. Avoid repeating completed tool actions when resuming an agent run. An uncertain generation result should remain uncertain until the application can verify its outcome.
What IsAnythingDown provides today
This version provides free server-rendered provider reports, source links, collection timestamps, component scope and recent collection attempts. Guides also have a Markdown version. OpenRouter is an official-page link without automatic collection. Jev uses the TypeSafe platform report rather than an OpenRouter route report.
There is no public diagnosis API, Model Context Protocol (MCP) server, automated model failover or independent AI inference probe. Local GitHub observations do not establish that an agent can reach an AI endpoint. Use the coverage page to inspect the checks that actually exist.
Related service pages
Each entry shows its own collection coverage. A linked service is not necessarily automatically collected.
Editorial review date applies to this guide. Live report collection times appear on the service pages.