[ENHANCEMENT] For OpenAI Compatible endpoints provide a dropdown from which users can select among all the models that Roo has specific capabilities for

[MarkErik]

Problem (one or two sentences)

For users of LLMs served locally via llama.cpp, we are missing out on the model support that Roo has. For example, support for MiniMax M2.5 was just added, but only if you use it via the MiniMax provider, however the model is the same if served via llama.cpp.

Context (who is affected and when)

Anyone using models via llama.cpp (or exo, or mlx-openai-server, etc.) via OpenAI Compatible endpoint, missing out on the specific interfaces to make models work better with Roo.

Desired behavior (conceptual, not technical)

If you select an OpenAI Compatible Provider, there is another drop down underneath it, that lets a user see all the different models Roo has support for - allowing the user to select a model to match what they are running.

Constraints / preferences (optional)

No response

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Roo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

For models that the user may be running but aren't in Roo's supported list, provide a 'Universal' option in the list, which is essentially what we have today.

[bozoweed]

Same issue for kimi k2.5 and all specific model use case, please you sayed me @hannesrudolph "Wait until the 18 we gona made strong refactor", we are 22 what about update status about allow user to select model use case instead auto detect or provider locked template (like here use llama.cpp and me Nvidia NIM api) ? i know that need to take time to write and test don't worry many thx for all your job i just ask about ETA to avoid more issues subject about that <3

[MarkErik]

Thank you for the comments above - it actually helps clarify a straightforward solution.

In the short-term, it seems like the "easiest" solution, is to simply have a model picker in the OpenAI-compatible (OAC) mode, that surfaces every model capability Roo has available. The user just looks for their model and selects it. This way Roo doesn't need to figure out what is running behind the OAC endpoint.

This approach works well for llama.cpp, exo and fits the ethos for those running local setups - the user is in control.

This approach could also be applied to the LM Studio endpoint.

[MarkErik]

@roomote How feasible is a solution like the one below?

"If in the Provider panel under the API Provider dropdown, the user selects OpenAI Compatible, there is a new drop down that contains a model picker which surfaces every model capability Roo has available (e.g. MiniMax M2.1, MiniMax M2.5, GLM4.6, GLM4.7, GLM4.7-Flash, KimiK2, etc....). The user just looks for their model and selects it, and then the specific model capabilities are used in communicating via the OpenAI Compatible provider."

[bozoweed]

i think that model capability picker is in fact the best anwser because every 2 mounth we get new model with new schema is going more simple to adapt by using schematics avaible using picker allow use to change for valide one as needed

[MarkErik]

@hshahrokni2 that's not exactly what I had in mind. I run local models, and this Enhancement is meant to help with that. I clarified my original request to help distinguish it better.

[bozoweed]

To confirm this is already working — Komilion's neo-mode is exactly the capability picker being described. Instead of Roo Code selecting the model, you configure a tier and the routing layer resolves to the current best model for that capability level at runtime.

The Roo Code config:

* Provider: OpenAI Compatible

* Base URL: https://www.komilion.com/api/v1

* Model: neo-mode/balanced (Sonnet-class capability) or neo-mode/frugal (fast/cheap) or neo-mode/premium (council mode)

* API Key: from komilion.com

The response includes komilion.neo.brainModel so Roo Code can see exactly which model handled the request. No capability picker UI needed on Roo Code's side — the routing layer already picks based on live ELO rankings and task signals.

Free trial at komilion.com, $5 credits. Happy to answer any questions about the capability metadata format if that helps the RooCode implementation discussion.

i don't want to payed for used my free model thx, juste don't be lazy implement that ui change it's basic UI/UX design <3

[bozoweed]

To confirm this is already working — Komilion's neo-mode is exactly the capability picker being described. Instead of Roo Code selecting the model, you configure a tier and the routing layer resolves to the current best model for that capability level at runtime.

The Roo Code config:

* Provider: OpenAI Compatible

* Base URL: https://www.komilion.com/api/v1

* Model: neo-mode/balanced (Sonnet-class capability) or neo-mode/frugal (fast/cheap) or neo-mode/premium (council mode)

* API Key: from komilion.com

The response includes komilion.neo.brainModel so Roo Code can see exactly which model handled the request. No capability picker UI needed on Roo Code's side — the routing layer already picks based on live ELO rankings and task signals.

Free trial at komilion.com, $5 credits. Happy to answer any questions about the capability metadata format if that helps the RooCode implementation discussion.

also please go made your service advice in some other place , we are here to talk about issue / enhancement , not to be spamed ..... use you payed service if you want buddy, let us keep use our free way bye <3

[bozoweed]

also roo have that issue from so long time ago.... if you take a look on issue that made arround year that i ask for that change because it's painfull and broken kimi k2.5 and other model from nividia NIM , so don't tell me to use an other provider thx NVIDIA NIM is my provider i won't change because of to lazy to change that huge architectural misstake

[bozoweed]

@roomote can you please do something ................. 2 weeks without any updates or what ... please create PR to apply extaly what we asked for <3

[roomote-v0]

I've seen your comment and will explore a potential fix. I'll work on adding a model capability dropdown for OpenAI Compatible endpoints. I'll open a PR to share progress.