feat(providers): add Requesty as an OpenAI-compatible provider

[Thibaultjaigu]

All Submissions:

• Notify authors: * @agentforce314, @ericleepi314

New Feature Submissions:

• Have you added an explanation of what your changes do and why you'd like us to include them?

Description

Adds Requesty, an OpenAI-compatible gateway to many model vendors, as a provider. It is a single row in the ProviderSpec registry, like the other hosted OpenAI-compatible vendors, and it uses hybrid catalog mode like OpenRouter: the curated list comes first and the live list is appended. It is opt-in only. It is not the default and is not part of any fallback order.

Changes proposed in this pull request:

• src/providers/openai_compatible_specs.py: new requesty row with base URL https://router.requesty.ai/v1, key from REQUESTY_API_KEY, and default anthropic/claude-sonnet-4-5 (the same default OpenRouter uses). The curated list is seeded from Requesty's managed model ids. All ids were checked against the live catalog.
• src/providers/model_discovery.py: new requesty discovery kind. It reads GET /models/managed (keeps only chat models) and falls back to the standard /models catalog when that fails. The existing kinds are unchanged.
• Tests: registry expectations, env var mapping, and two parser tests (managed list preferred, catalog fallback).
• Desktop settings: a REQUESTY_ key group. VS Code extension: label and env var entries next to OpenRouter.
• Docs: README provider list, SETUP_GUIDE table row plus a short note (the EU endpoint https://router.eu.requesty.ai/v1 works through base_url), and a CHANGELOG entry.

How to test:

export REQUESTY_API_KEY=...
clawcodex --provider requesty --model openai/gpt-4o-mini

Checks run locally:

• pytest tests/test_provider_registry.py tests/providers/ tests/test_opencode_compat_providers.py tests/test_providers.py: 379 passed
• VS Code extension npm test: 84 passed, node scripts/lint.js clean
• Live call through get_provider_class("requesty") with openai/gpt-4o-mini: 200, the reply came back and the model was echoed as gpt-4o-mini-2024-07-18. Discovery returned 159 managed models, and the hybrid merge kept the curated ids first.
• The desktop typecheck was not run locally (it needs a full npm ci). The change there is one object in PROVIDER_GROUPS that uses the same shape as its neighbours.

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