Add Requesty provider

[Thibaultjaigu]

Summary

Adds an optional direct Requesty provider route to the SDK. Requesty is an OpenAI-compatible model router (https://router.requesty.ai/v1) that uses the same provider/model id form as OpenRouter (e.g. openai/gpt-4o-mini, anthropic/claude-sonnet-4.5).

When the REQUESTY_API_KEY environment variable is set, chat completions are routed directly to the Requesty router using the existing @codebuff/llm-providers OpenAICompatibleChatLanguageModel shim, instead of going through the Codebuff backend. When the env var is unset, behavior is unchanged.

This mirrors the existing direct ChatGPT OAuth route in sdk/src/impl/model-provider.ts (createOpenAIOAuthModel), which already builds an OpenAICompatibleChatLanguageModel pointed at a fixed upstream base URL with a Bearer key rather than the Codebuff backend. No new provider registry was invented — the change reuses the existing model-routing entry point (getModelForRequest).

Changes

Code:

• common/src/constants/requesty.ts (new): REQUESTY_BASE_URL and REQUESTY_ENV_VAR constants, mirroring common/src/constants/byok.ts.
• sdk/src/env.ts: getRequestyApiKeyFromEnv() helper, mirroring getByokOpenrouterApiKeyFromEnv().
• sdk/src/impl/model-provider.ts: createRequestyModel() (mirrors createOpenAIOAuthModel()) and a direct-route branch at the top of getModelForRequest().

Tests:

• sdk/src/impl/__tests__/model-provider-requesty.test.ts (new): asserts the route is selected when REQUESTY_API_KEY is set (provider === 'requesty') and that the default Codebuff backend route is used when it is unset.

Docs:

• sdk/README.md: new "Model providers" section documenting the Requesty route.
• docs/agents-and-tools.md: matching "Model providers" section.

Verification

Typecheck (no repo-wide script; ran the affected projects):

• sdk: tsc --noEmit -p tsconfig.json → exit 0
• packages/llm-providers: bun run typecheck → exit 0

Tests (bun test):

• model-provider-requesty.test.ts (2) and model-provider-free-mode.test.ts (4) → 6 pass / 0 fail.

Live test against the Requesty router:

• GET https://router.requesty.ai/v1/models → HTTP 200
• POST https://router.requesty.ai/v1/chat/completions with openai/gpt-4o-mini → real completion returned (model gpt-4o-mini-2024-07-18).

Notes

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