Text-only models now have vision in Command Code. • A vision model reads the attached image • Your main model gets the context in the same thread • /config lets you choose the vision model Stop describing screenshots. Just attach them.

This is really great.. Waiting for the desktop app release.

tbh one of the most GOATED features of Command Code.

Now this is goated fs

command code is goated Can we more free models pleaseeee

Laguna is free

atleast someone thought of actually adding it

🐐

alright JEEZ V4 Flash ain't gonna need modlens anymore :))

nops.

Chef's kiss for this feature. Super impactful.

Nice leap.

Needed feature

You should open source the desktop app so we can help you get it faster

it's almost ready!

@CommandCodeAI why did this MUST have feature take soo much time guys !!😅

it's been there for a month really. we're bad at sharing things since we ship more than we can share lol.

cool, deepseek v4 flash now has👀

We need an official extension for vscode and zed

Soon!

♥️.

Now this is one of the best additions in value. Goated PM who thought of this!!

okay.. here is the issue with this.. I really love your stuff, so I am saying this purely out of respect and hoping that you guys would resolve this LONG life issue I have... auxiliary models for vision NEVER works... PURELY due to information gets LOST when attempt to be described.. if I share an image saying "this UI is terrible, please fix it" the auxiliary vision model would see the image, would describe its OWN version of the issue and then sends it to the main model.. however, the main model would have the image AS REFERENCE as pure vision in its thinking, were it can compare things as it re-tries. so if I say match the colour and it guesses the wrong colour, it can STILL review the image again and see if it matches. An auxiliary model cant do this.. its a one time thing.. my solution would be a unique skill that converts images from pixels into some sort of blob, that the model would then be trained to see it.. so it can figure out what the blob is.. this way it can "see" the image as text and it would be pixel perfect vision. happy to be corrected here if I got my wires mixed up!

so can we use this feature with different apps? using commandcode just for backend and claude cli or codex etc @CommandCodeAI

This is a really clean way to bolt vision onto text-only models.

Guardar esa lectura de la imagen en el hilo evita que el modelo de visión la relea en cada turno y mantiene plano el gasto de tokens.

hey can you make the vision model uncensored or the ability to switch it with one ?

you can switch. and byok will have you switch to literally anything.

Guys the "Usage" section in your web is broken, I cant visualize my usage per model since yesterday. Can you fix this?? @CommandCodeAI

How many Deepseek v4 flash requests per 5 hour limit/monthly limit, on the $10 goat plan? I keep seeing posts saying not to trust the numbers. I'll buy the sub right now if it can be confirmed that it's far above OpenCode Go, while still using the 0731, non quant model

@opencode feature request? This would be dope.

goat'd fr