Add Requesty LLM router as a pricing source

[BillJr99]

Summary

Adds support for Requesty, an LLM router service, as a new pricing source. This enables automatic detection of available models and their pricing from Requesty's OpenAI-compatible /v1/models endpoint.

Key Changes

• New source implementation (scripts/sources/requesty.py):

  • Fetches models from Requesty's /v1/models endpoint using Bearer token authentication
  • Parses flat pricing fields (input_price, output_price) at the top level
  • Correctly identifies genuinely free models (zero input/output cost) vs. paid models
  • Handles missing/invalid pricing data gracefully by treating as infinity
  • Prefixes model IDs with "requesty/" for fully-qualified identification

• Provider configuration (llmproxy/providers.json):

  • Added Requesty provider entry with base URL and API key requirement
  • Configured with display name and key hint pointing to app.requesty.ai

• Configuration templates (config.example.json):

  • Added example Requesty configuration block with base URL and API key placeholder

• Documentation (README.md):

  • Updated sources table to include Requesty as a medium-confidence source

• Comprehensive test coverage (tests/test_scraper/test_requesty_source.py):

  • Tests free model detection (zero pricing → is_free=True)
  • Tests paid model detection with per-token pricing extraction
  • Verifies API key requirement (no HTTP call without REQUESTY_API_KEY)
  • Validates high confidence rating and proper source attribution
  • Confirms Bearer token authentication header is sent

• Test fixtures (tests/fixtures/requesty_models.json):

  • Sample response data with mix of free (Nemotron, Gemma) and paid (Claude, GPT) models

Notable Implementation Details

• Requires REQUESTY_API_KEY environment variable; silently returns empty list if not set
• Pricing is stored as per-token dollar amounts (consistent with OpenRouter convention)
• Free models have pricing=None to avoid expressing a pricing opinion
• Robust error handling for malformed responses and missing fields
• High confidence rating due to direct API source with explicit pricing metadata

https://claude.ai/code/session_01BxBkHZLhDB8aqrJVaYkmht