[Feature]: expose requestPacing.maxConcurrentRequests in the provider settings GUI #6093

What are you trying to accomplish?

Cap how many requests to a provider are in flight at once — provider-wide, with a tighter per-model limit — from the dashboard, the same way RPM and minimum interval are managed today.

What prevents this today?

requestPacing.maxConcurrentRequests shipped in the backend via #5954 (schema in src/config/schema/leaf-validators.ts, enforcement in src/providers/request-pacing.ts and src/server/responses/fetch-helpers.ts), but the provider workspace pacing panel only exposes enabled, requestsPerMinute, minIntervalMs, and per-model RPM/interval overrides. The concurrency cap is reachable only by editing the config file or calling the management API directly. #5708 carried this GUI work and was closed as superseded, while #5954 landed backend-only.

What should OpenCodex do?

The Request pacing card in Dashboard → Providers → → Settings should offer a "Max concurrent requests" input at the provider level and a cap field in each per-model override row, following the landed validation: a positive integer, and a pacing rule may consist of the cap alone.

Example usage or interface

Today the cap is config-only:

"requestPacing": {
  "enabled": true,
  "requestsPerMinute": 120,
  "maxConcurrentRequests": 5,
  "models": { "some/model": { "maxConcurrentRequests": 2 } }
}

After this change the same values are editable (and visible) in the pacing panel next to RPM and minimum interval.

Alternatives or workarounds

Edit ~/.opencodex/config.json by hand, or PATCH /api/providers?name=<provider> from a terminal. Both work but leave the dashboard silently out of sync with the effective configuration.

Additional context

• Original pacing proposal: [Feature]: requestPacing option to cap concurrent in-flight requests at provider and model level #5702

• Backend implementation: feat(proxy): support maxConcurrentRequests in requestPacing at provider and model level #5954 (via merge train feat: integrate non-GUI enhancement batch 10B #5992)

• The closed feat(providers): requestPacing.maxConcurrentRequests at provider and model level #5708 contained this GUI work before the backend landed separately.

• I searched existing issues and documentation.

• This request describes a concrete OpenCodex workflow rather than merely naming a desired technology.

• I removed secrets and personal data.