# Rate limits and versioning

Read rate-limit headers and error bodies, use bounded exponential backoff with jitter for retryable responses, cap concurrency, and pin the documented API version behavior.

## Where this lives

Use a trusted local or server-side API client.

The Foxora API provides OpenAI-compatible chat completions plus account, usage, and public model-discovery endpoints. Authenticated routes currently accept a signed-in Foxora account access token.

> **Safety:** Treat the complete account access token as a secret. Do not paste it into prompts, commit it to a repository, expose it in client-side bundles, or include it in logs, screenshots, and support messages.

## Steps

1. Use a trusted local or server-side API client.
2. Read rate-limit headers and error bodies.
3. Use bounded exponential backoff with jitter for retryable responses, cap concurrency.
4. Pin the documented API version behavior.
5. **Confirm the result.** Inspect HTTP status, response headers, request identifiers, usage fields, and the parsed response; cross-check account or usage endpoints after a small request.

## Choose the right path

- Use `/v1/chat/completions` for OpenAI-compatible generation.
- Use `/v1/me` and `/v1/me/usage` for account and usage state.
- Use `/v1/models` to discover model identifiers instead of hardcoding assumptions.


## Confirm it worked

- Inspect HTTP status, response headers, request identifiers, usage fields, and the parsed response; cross-check account or usage endpoints after a small request.
- The screen, command, file, run, or destination named in this guide reflects the expected state.