[Feature]: Manual Override for Model Capabilities

[NexiaMoe]

1. Why Do You NEED This Feature?

Currently, ForgeCode derives model capabilities (e.g., image, video, tools support) directly from the LLM provider. In practice, this metadata is not always accurate or complete. Some models that actually support certain features are incorrectly marked as unsupported, which limits their usability within the CLI.

Benefits:

• Enables advanced users to fully utilize model capabilities
• Avoids dependency on potentially inaccurate provider metadata
• Improves flexibility for self-hosted or proxy-based LLM setups
• Future-proofs ForgeCode against inconsistencies across providers

2. What Is NOT Possible Right Now?

Right now, Like kimi model, it should support Image capability, but cannot be override currently

3. What WILL Be Possible With This Feature?

• Use models with full capabilities even when the provider metadata is incorrect or incomplete
• Enable image input/output for models that actually support vision but are reported otherwise
• Enable tool/function calling for models that support it but are flagged as unsupported
• Work seamlessly with self-hosted or proxied LLMs where capability metadata is missing or unreliable
• Customize model behavior to fit real-world usage instead of provider assumptions
• Avoid being blocked by provider-side limitations or delayed updates in capability reporting

Example scenarios:

• A user enables image support for a model behind a proxy that doesn’t expose capability metadata
• A locally hosted model (e.g., via Ollama or custom backend) can use tools after manual override
• Advanced users fine-tune their workflow by enabling only the features they actually need

Proposed Solution (User Experience)

Introduce a manual override mechanism that allows users to define or adjust model capabilities locally. This could include:

1. Overriding support for:

• Image input/output
• Video input/output
• Tool/function calling

2. Defining capabilities via a config file (e.g., .forge.toml or similar)
3. Merging provider-reported capabilities with user-defined overrides (user config takes precedence)

Alternatives Considered

No response

Feature Category

AI/LLM Integration

Priority/Impact

High - Would significantly improve my workflow

Examples from Other Tools

No response

Additional Context

No response

Pre-submission Checklist

• I have searched existing issues and confirmed this is not a duplicate
• I am willing to submit a PR to implement this feature

[amitksingh1490]

@NexiaMoe https://forgecode.dev/docs/custom-providers/#static-model-list does this fix this issue?

[NexiaMoe]

Yup, adding input_modalities should be fix the issue, but does it need [[providers]] ? Is it possible to only override the models if the model id is match ? So It will not to override the provider url and key as it included to the program

[joshuas99]

I second this request.

Separately but related, there may be a bug in the static model list capability. I have attempted to use it as per the documentation linked and could not get the toml configuration to be accepted. Because of this I can't set my local model context size, which is quite frustrating.