Vision for text-only models
Command Code provides text-only models with a unique ability to read images in the session. When you paste a screenshot, diagram, or error dialog, a text-only model has no way to see it.
Vision capabilities close that gap. When you attach an image to a text-only model, Command Code tells the model an image is there, and the model can read it on demand with the VISION tool, a single side-call to a cheap, vision-capable model that transcribes the image to text.
You can enable/disable the vision tool, and also select a vision capable model of your choice from /config menu.
Vision capability is opt-in. The first time the model tries to read an attached image, Command Code asks whether to enable the feature globally.
You can change it any time from /config. The Image vision row cycles between Ask on first use, Enabled, and Disabled, resetting it restores "Ask on first use", so the consent prompt returns the next time the model reads an image.
| Setting | Value |
|---|---|
| Vision model | MiMo V2.5 (default) |
| Enable / disable | /config → Image vision |
| Change vision model | /config → Feature models → Image vision |
| Billing | The side-call bills as normal usage, surfaced to you |
The vision call is independent of your main session model, so it does not change which model answers your prompt. Each call is bounded by a timeout and fails open: if the vision model is unavailable, the turn continues and the VISION row shows the failure reason so it stays diagnosable.
- Vision is only invoked for models that cannot read images natively. Vision-capable models receive the image directly — no markers, no tool.
- If vision is disabled, attaching an image to a text-only model shows an advisory suggesting you turn on image vision in
/config, or switch to a vision-capable model with/model. - The model can also read image files: a
read_fileof an image path makes that image inspectable with theVISIONtool.