feat: add Requesty as an OpenAI-compatible upstream

[Thibaultjaigu]

Add Requesty as a supported OpenAI-compatible upstream

This adds Requesty as a first-class OpenAI-compatible upstream in codex-provider-relay, mirroring the existing OpenRouter provider as closely as possible. Requesty is an OpenAI-compatible LLM gateway (base URL https://router.requesty.ai/v1, Authorization: Bearer <key>, provider/model naming such as openai/gpt-4o-mini), so it slots into the same generic adapter path without any new provider plugin class.

What changed (mirrors OpenRouter at every wiring site)

• Capability preset (packages/codex-provider-relay/src/capabilities/capability_presets.ts):
  • Added 'requesty' to the OpenAICompatibleCapabilityPresetId union.
  • Added a requesty preset to PRESETS (apiKeyEnv: REQUESTY_API_KEY, baseUrl: https://router.requesty.ai/v1, defaultModel: openai/gpt-4o-mini, ownedBy: requesty).
  • Added a { presetId: 'requesty', envPrefix: 'REQUESTY' } entry to OPENAI_COMPATIBLE_PROFILE_PRESET_REGISTRATIONS so it is loaded from env exactly like the other presets.
  • Added 'requesty' to the normalizePresetId switch.
• Model catalog (packages/codex-provider-relay/src/capabilities/cliproxy_model_catalog.ts): added 'requesty' to CliproxyModelCategory and a requesty / openai/gpt-4o-mini catalog entry mirroring the OpenRouter one.
• README smoke env group (README.md): added REQUESTY_API_KEY / REQUESTY_BASE_URL / REQUESTY_DEFAULT_MODEL to the "Supported smoke env names" block, plus the provider listings and the CODEX_COMPAT_CAPABILITIES / optional-keys references.
• Smoke env consumer (test/providers/openai_compatible/live_provider_smoke.test.ts): added a requesty entry to the live PROVIDERS list so the new env group is actually consumed (the preset registration drives REQUESTY_* env loading via loadCodexProfilesFromEnv), not docs-only.
• Recipes (packages/codex-provider-relay/docs/RECIPES.md, LIVE_SMOKE_RECIPES.md): added a parallel Requesty CodexProviderRuntime recipe (upstreamBaseUrl: "https://router.requesty.ai/v1", providerLabel: "requesty", apiKey: process.env.REQUESTY_API_KEY, defaultModel: "openai/gpt-4o-mini"), a requesty capability-preset server-env example, and a Requesty live-smoke env block. The OpenRouter examples are unchanged.

Verification

• tsc -p packages/codex-provider-relay/tsconfig.json --noEmit — clean (0 errors).
• tsx --test packages/codex-provider-relay/test/*.test.ts — 181 pass, 1 skip, 0 fail.
• Root tsc --noEmit error count is identical before and after this change (pre-existing, unrelated test-file type issues only); my edited files introduce no new errors.
• Confirmed via the real loader code path that getOpenAICompatibleProviderPreset('requesty') resolves to baseUrl https://router.requesty.ai/v1, apiKeyEnv REQUESTY_API_KEY, defaultModel openai/gpt-4o-mini, and that the REQUESTY registration is present.
• Live chat completion against https://router.requesty.ai/v1/chat/completions with openai/gpt-4o-mini returned HTTP 200 with a real completion, confirming the documented base URL and model naming work.

Docs: https://requesty.ai , https://docs.requesty.ai

I work at Requesty. This mirrors the existing OpenRouter provider as closely as possible. Happy to adjust or close it if it's not a fit.