hermes - 💡(How to fix) Fix Feature Request: Model allowlist for ollama-cloud provider [1 pull requests]

Problem

The ollama-cloud provider uses live dynamic discovery — it hits ollama.com/v1/models on every /model invocation, merges with models.dev, and shows everything. For users with a small, known set of models (5 in my case), this means scrolling through a long list of irrelevant models every time.

Providers with curated catalogs (OpenRouter, Nous Portal) get clean picker lists. Dynamic providers like ollama-cloud and lmstudio don't have an equivalent filtering mechanism.

Proposed solution

Add a model_allowlist config key to model_catalog.providers.<provider> that lets users specify which models appear in the picker:

model_catalog: providers: ollama-cloud: model_allowlist: - deepseek-v4-pro - gemma4:31b - glm-5.1 - qwen3.5 - minimax-m2.7

When set, the picker shows only that subset. When empty (default), behavior is unchanged — show all dynamically discovered models.

Affected code

• hermes_cli/config.py — model_catalog.providers default schema
• hermes_cli/model_switch.py — get_model_picker_entries(), around line 1171 where fetch_ollama_cloud_models() feeds the curated dict
• Could be generalized to apply to any dynamic-discovery provider (ollama-cloud, lmstudio, custom providers that return large lists)

Alternatives considered

• Self-hosting a curation JSON and pointing model_catalog.providers.ollama-cloud.url at it — works but adds infrastructure overhead
• Freezing the cache file — fragile, breaks on TTL expiry
• Skipping the picker and using /model <name> directly — works but discards the picker UX