feature: dont bundle agents/models/providers directly into code, but use the `.openfang` folder to store

[9M6]

Description

If we need a clean install without any providers, models, agents etc, this is currently difficult to achieve, since I believe the defaults are embedded into the binary, it should be a better practice to put everything into .openfang to allow customization, or the ability to do a clean openfang install without anything and customize it myself.

Alternatives Considered

N/A

Additional Context

No response

[jaberjaber23]

OpenFang already supports custom models via POST /api/models/custom and custom provider URLs via PUT /api/providers/{name}/url. Agent templates are loaded from ~/.openfang/agents/ directory on disk. The bundled defaults exist as compile-time fallbacks but disk configs take precedence. For fully externalizing the model catalog and provider list, this would require replacing the include_str!() bundled data with file-based loading. Tracked for future consideration alongside #545.

[jaberjaber23]

Reopening — this is a valid architectural concern about externalizing bundled agents, models, and providers from include_str!() into configurable files in ~/.openfang/. Disk configs already take precedence but the bundled defaults are still compiled in. Leaving open for proper design consideration.

[jaberjaber23]

Agents are already partially externalized — bundled templates are installed to ~/.openfang/agents/ on init, and disk versions take priority over compiled-in ones. Custom models load from ~/.openfang/custom_models.json. Providers are still hardcoded in the binary. Full externalization (loading models and providers from disk) is a larger refactor that would touch model_catalog.rs, kernel boot, and CLI init. Leaving open as a tracked improvement.