feat: fetch provider models dynamically and fix isOpenSource pollution

[RomneyDa]

Summary

1. Fixes issue where isOpenSource is used to detect if a model can be used by ollama. Changes to explicit providers listing.
2. Adds support for dynamically fetching provider models. Open router already has this, moves logic to core to unify. Adds support for ollama using their tags API. Adds support for other providers using their listModels methods (or custom for gemini/anthropic). A little refresh icon to "update models list using my API key" shows once an API key has been entered. Does some sensible filtering and model param detection for context length, max tokens, and tool use. Also these results are deduplicated with hardcoded models (hardcoded take precedence) and show proper provider icons etc.
3. Adds explicit Gemma 4 and GPT 5.4 family support
4. updates some tools support regex matching to match e.g. gpt 6, 7, etc.

---

Summary by cubic

Unifies dynamic model fetching across providers via a core models/fetch protocol, auto-loading ollama and openrouter, and letting users refresh other providers with their API key. Persists fetched context, max tokens, and capabilities to config.yaml, adds GPT‑5.4 (pro/mini) and Gemma 4, and removes isOpenSource-based routing.

• New Features

  • Central models/fetch: ollama scrapes the library (icons), openrouter hits its public API on load, anthropic calls /v1/models, gemini calls v1beta/models (chat-only; excludes 2.0, Gemma, embeddings/TTS/robotics/etc.), others use listModels() with optional apiKey/apiBase; returns contextLength, maxTokens, supportsTools; errors toast and return [].
  • Add Model: Auto-loads ollama/openrouter; refresh button appears after API key to fetch provider models; merges and de‑duplicates into “Additional models”; sorts fetched models.
  • Persistence: Writes contextLength, completionOptions.maxTokens, and capabilities (tool_use, image_input) to config.yaml when adding a model.
  • Models: Adds gpt-5.4 (pro, mini) and gemma4; improves image/tool detection for Gemma 4 and future GPT 5+.

• Bug Fixes

  • OpenAI.listModels(): Supports providers returning either a plain array or { data: [] }; error text now references config.yaml.
  • Add Model UX: Centered spinner; clears API key when switching to providers without RequiresApiKey and after adding; selects the new model by posting directly to core.
  • Routing: Replaces isOpenSource logic with explicit providerOptions and popularPackages.
  • Tool/agent heuristics: Safer regexes for GPT 5+ (gpt-[5-9]), Claude 4+, Grok 4.x+, GLM 4+; OpenAI checks use isOSeriesOrGpt5PlusModel.

^{Written for commit bcd29eb. Summary will update on new commits.}

[cubic-dev-ai]

3 issues found across 12 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="gui/src/forms/AddModelForm.tsx">

<violation number="1" location="gui/src/forms/AddModelForm.tsx:73">
P2: Late async model-fetch responses can overwrite state after provider switch, showing stale models from the previous provider.</violation>
</file>

<file name="core/llm/fetchModels.ts">

<violation number="1" location="core/llm/fetchModels.ts:100">
P2: Ollama model discovery relies on brittle scraping of `x-test-*` HTML markers, which can silently return empty/partial model lists when upstream markup changes.</violation>
</file>

<file name="gui/src/pages/AddNewModel/configs/providers.ts">

<violation number="1" location="gui/src/pages/AddNewModel/configs/providers.ts:77">
P2: Ollama catch path resets `packages` but not `popularPackages`, allowing stale popular-model state to persist across re-initializations.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.}

[cubic-dev-ai]

You're iterating quickly on this pull request. To help protect your rate limits, cubic has paused automatic reviews on new pushes for now—when you're ready for another review, comment @cubic-dev-ai review.