feat: add Requesty as an OpenAI-compatible provider

[Thibaultjaigu]

Summary

Adds Requesty as a named chat provider. Requesty is an OpenAI-compatible LLM gateway, so it is wired next to OpenRouter using the existing _OpenAICompatStreamer.

Changes

• api/clients/requesty.py: RequestyClient on top of adalflow's OpenAIClient. Uses REQUESTY_API_KEY, defaults to https://router.requesty.ai/v1 and honours an optional REQUESTY_BASE_URL (for example https://router.eu.requesty.ai/v1 for the EU region). Sends the same HTTP-Referer / X-Title headers as the OpenRouter client, and sends a plain text prompt as a user message like the OpenRouter client does (a lone system message is rejected by Anthropic and Gemini models).
• api/chat/_stream.py: RequestyChatStreamer registered as requesty.
• api/clients/__init__.py, api/config.py: export, env var, CLIENT_CLASSES and default provider map entries.
• api/config/generator.json and generator.litellm.json: requesty provider block with a few models (default openai/gpt-4o-mini), custom models allowed. The default provider is unchanged.
• src/messages/*.json: providerRequesty label for the provider selector.
• api/README.md: env vars, key link and provider list.
• Tests: requesty added to the registry tests in tests/unit/test_chat.py, plus tests/unit/test_requesty_client.py for base URL handling and message conversion.

How to test

REQUESTY_API_KEY=rqsty-...

Start the backend, pick "Requesty" in the model selector and use openai/gpt-4o-mini (or any custom model id such as anthropic/claude-sonnet-4-5). Keys are at https://app.requesty.ai/api-keys.

Validation

• ruff check and ruff format --check (v0.8.6, as pinned in pre-commit) pass on all touched Python files.
• pytest tests/unit tests/backend tests/test_config.py: 81 passed, 10 failed. The same 10 fail on a clean main in my environment (missing GOOGLE_API_KEY / OPENAI_API_KEY and watchfiles), and the new tests all pass.
• Ran the real stream through ChatStreamer.create(provider="requesty", ...) with the config from get_model_config, and every model in the new block replied: openai/gpt-4o-mini, openai/gpt-4.1, openai/gpt-5-mini, anthropic/claude-sonnet-4-5, google/gemini-2.5-flash, deepseek/deepseek-chat.
• The frontend change only adds a translation key; JSON files validated, I did not run the Next.js build.

Disclosure: I work at Requesty. Happy to adjust anything to match project conventions.

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.