1. Create a Cursor identity
In your SafetyRouter tenant, create a separate API key for the employee's Cursor installation. Add team and project tags so its activity appears in the right reports.
2. Open Cursor model settings
In Cursor, open Cursor Settings → Models. Find the OpenAI API-key settings and enter the SafetyRouter key:
sr_live_...3. Override the OpenAI base URL
Enable Override OpenAI Base URL and enter your tenant endpoint:
https://your-company.safetyrouter.io/v1For an office-only installation, use the reachable LAN address instead:
http://192.168.166.105:8000/v14. Select an allowed standard model
Choose a standard, non-reasoning OpenAI chat model whose exact identifier is enabled in SafetyRouter. Click Verify, then send a simple Chat request and confirm it appears in the tenant request log.
Compatible Cursor Chat requests pass through SafetyRouter's input policy, output inspection, rate limits, model allowlists, budget controls, tags, and audit trail.
Agent: preview only. Some current Cursor Agent flows use Responses-style payload fields even when calling /chat/completions; SafetyRouter does not yet promise full Agent compatibility.
Tab/autocomplete: not routed through SafetyRouter. Cursor documents that specialized features continue using Cursor's built-in models.
Base URL override: the override can affect all OpenAI-family selections. Disable it when returning to Cursor-managed OpenAI models.
Roadmap
- Normalize and test Cursor Agent's Responses-style request variants.
- Verify tool-call loops and buffered-safe streaming across supported Cursor releases.
- Publish a versioned integration test matrix before labeling Agent support as tested.
BYOK scope verified against the official Cursor API Keys documentation. Current override and Agent limitations are tracked in the Cursor support forum.