Summary

This PR establishes golemcore-models as a shared, provider-agnostic registry of model defaults for the Model Catalog discovery flow in golemcore-bot.

The repository now contains curated ModelSettings JSON files for current, general-purpose text and vision models across OpenAI, Anthropic, and Gemini. The intent is not to mirror every vendor model ID, but to provide stable default configurations for models that are expected to be selected interactively from discovery results.

What this adds

• a stable shared registry layout based on models/<model-id>.json
• provider-agnostic ModelSettings payloads with no provider field
• shared defaults for the current model families used in catalog creation flows:
  • OpenAI GPT-5 family: gpt-5.1, gpt-5.2, gpt-5.2-pro, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro
  • Anthropic Claude family: claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5
  • Gemini family: gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite, gemini-3-flash-preview, gemini-3.1-flash-lite-preview, gemini-3.1-pro-preview
• repository-level documentation in README.md describing layout and lookup order

Scope and intent

This registry is intentionally opinionated.

It is designed to support default config resolution during model creation, so it includes only current, general-purpose models that make sense in a shared catalog. It explicitly avoids turning the repository into a full upstream mirror of every vendor endpoint or snapshot.

As a result, the catalog excludes:

• deprecated models
• dated aliases and snapshot-style IDs
• specialized endpoint-only models such as audio, realtime, image-generation, and other non-catalog artifacts

Notes

• OpenAI GPT-5 family entries are modeled conservatively around current parameter compatibility, especially for temperature support.
• Reasoning-enabled entries include explicit reasoning defaults and level maps where that behavior is part of the model family contract.
• The repository layout matches the lookup contract already implemented in GolemCore: providers/<provider>/<model-id>.json first, then models/<model-id>.json.