Enhancement: Use models.dev as LLM model registry source

[yanurag-dev]

Summary

Adopt models.dev (https://models.dev) as the primary source for LLM model metadata instead of maintaining a static provider.json file. This would keep the model registry automatically up-to-date with pricing, capabilities, and context limits.

Context

Currently, forgecode maintains a static provider.json file (~3200 lines) with hardcoded model definitions. This requires manual updates when new models are released.

I analyzed how Dexto (another AI coding assistant) handles this - they sync from models.dev at build time.

Current Approach (forgecode)

• Static crates/forge_repo/src/provider/provider.json embedded at compile time
• Some providers dynamically fetch models from their APIs at runtime
• Manual updates required for new models
• No pricing information

Proposed Approach (use models.dev)

• Fetch model catalog from https://models.dev/api.json at build time
• Generate a Rust file similar to Dexto's models.generated.ts
• Automatically includes:
  • Model IDs and names
  • Context limits (input/output tokens)
  • Pricing (input/output per million tokens, cache pricing, reasoning pricing)
  • Modalities (text, image, audio, video, pdf)
  • Capabilities (reasoning, tool_call, temperature support)
  • Release dates and status

Benefits

1. Always up-to-date - New models automatically available after running sync script
2. Rich metadata - Pricing, modalities, and capabilities come free
3. Less maintenance - No manual updates needed for model list
4. Cross-provider - Single source for all providers (OpenAI, Anthropic, Google, etc.)

Implementation Suggestion

Create a scripts/sync-llm-registry.ts (or .rs) that:

1. Fetches https://models.dev/api.json
2. Filters models by provider-specific rules (e.g., OpenAI → gpt-*, o*)
3. Generates a Rust file with MODELS_BY_PROVIDER constant
4. Add to CI to verify registry is up-to-date

Reference

• Dexto's implementation: scripts/sync-llm-registry.ts
• Dexto's generated file: packages/core/src/llm/registry/models.generated.ts

Todo

• Create sync script
• Generate initial model registry
• Update existing provider code to use new registry
• Add CI check to verify registry is up-to-date