feat: add Requesty as an OpenAI-compatible provider

[Thibaultjaigu]

Add Requesty as an OpenAI-compatible provider

This adds Requesty as a first-class provider, mirroring the existing OpenRouter integration. Requesty is an OpenAI-compatible LLM gateway: it speaks the standard Chat Completions API, uses provider/model naming, and routes to the same upstream model providers.

Provider shape

• SDK: openai (openai_compatible: true)
• Base URL: https://router.requesty.ai/v1
• Auth: REQUESTY_API_KEY (sent as Authorization: Bearer <key>)
• Model naming: provider/model (e.g. openai/gpt-4o-mini, anthropic/claude-sonnet-4-5)
• Output token param: max_tokens

Wiring sites (mirrors OpenRouter exactly)

• pantheon/repl/setup_wizard.py — new ProviderMenuEntry("requesty", "Requesty", "REQUESTY_API_KEY", "REQUESTY_API_BASE") in PROVIDER_MENU, right after OpenRouter (same arg order: id, display, api_key_env, api_base_env).
• pantheon/utils/llm_catalog.json — new sibling "requesty" provider block matching the OpenRouter field shape (display_name, sdk, base_url, api_key_env, openai_compatible, chat_output_token_param, models). Catalog-driven resolution in llm_providers.detect_provider / provider_registry picks this up with no per-provider branching. Includes a small set of known-live Requesty models: openai/gpt-4o-mini, openai/gpt-4o, anthropic/claude-sonnet-4-5, deepseek/deepseek-chat, google/gemini-2.5-flash.
• pantheon/utils/vision_capability.py — added "requesty" to _KNOWN_CHAT_COMPLETIONS_ONLY alongside openrouter (gateway proxies Chat Completions, so tool-result images fall back to the sub-agent path, conservatively matching OpenRouter behaviour).
• pantheon/utils/model_selector.py — added requesty to DEFAULT_PROVIDER_PRIORITY, a requesty entry in DEFAULT_PROVIDER_MODELS (high/normal/low quality chains), and "requesty": "REQUESTY_API_KEY" in PROVIDER_API_KEYS.

Verification

• JSON validated: python3 -c "import json; json.load(open('pantheon/utils/llm_catalog.json'))" passes; requesty present with all 5 models.
• AST parse of every edited .py (setup_wizard.py, vision_capability.py, model_selector.py) passes.
• Smoke import: from pantheon.utils.model_selector import DEFAULT_PROVIDER_MODELS, PROVIDER_API_KEYS resolves requesty correctly.
• Ran tests/test_model_selector.py tests/test_setup_wizard.py tests/test_llm_provider_config.py tests/test_provider_adapters.py: same pass/fail counts as the unmodified baseline (pre-existing failures are unrelated environment/test-drift; this change introduces no new failures and adds passing provider-parametrized cases).
• Live chat completion against https://router.requesty.ai/v1/chat/completions with openai/gpt-4o-mini, max_tokens: 16 → HTTP 200, real completion returned (gpt-4o-mini-2024-07-18).

Links

• 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.