Choosing models

Every assistant needs a model to think with. StellarFirm can pick one for you, or you can choose. This page explains where each choice applies, how to keep open models in front, and how to bring a model of your own.

At a glance#

You want to	Do this
Use a different model for one chat	Pick it in the model picker under the message box.
Set the models an assistant always uses	Open Settings, Models and choose Auto or Manual on that assistant's card.
Keep the models you like close	Star them in the picker. Starred models stay at the top and in the column on its left.
Prefer open models	Filter the picker by Open, star the ones you like, or rank open models in Manual.
Use a model you already run	Add it under Your own models in Settings, Models.

Where a choice applies#

There are two places to choose, and they do different things.

1. The picker under the message box is for that chat. It changes the model of whoever answers in that chat: the CEO assistant in the CEO's chat, or the assistant you are talking to in its own chat. It does not change any other assistant. When the CEO assistant hands work to the Coder or to Marketing, each of them keeps its own setup.
2. Settings, Models is for every chat. Each assistant has its own card, and so does the CEO assistant. Auto lets StellarFirm pick the best models for that role. Manual is a list you rank yourself: the first model that can run answers, and the next one takes over if it cannot. To set many assistants at once, use the bulk actions on that screen, such as using Auto for everyone or applying one setup to the assistants you select.

So picking a model in the CEO's chat changes only the CEO assistant's answers in that chat. To put the whole team on a model, set it in Settings, Models.

Use open models first#

Open models suit most everyday work: drafting, research, summaries, routine coding and planning. Keep the strongest models for hard problems and high stakes work. Some ways to lean on open models:

• Filter by Open. In the picker, the Open filter shows only open models. Until you star a model, the picker offers Show open models at the top.
• Pin them. Star a model in the picker, or press the shortcut shown at its foot, and it stays at the top of the list. The star column on the left of the picker jumps straight to them. Your stars are kept with your account, so they follow you across devices.
• Rank them in Manual. Put open models first in an assistant's Manual list, and keep a stronger model below them as the fallback.
• Green mode. Where your workspace offers it, green mode in Settings, Models makes Auto try light, efficient open models first and keeps thinking low.

Auto leans on the strongest models for now. If you prefer open models, a Manual list is how you say so.

Bring your own models#

Your models sit next to StellarFirm's in the same picker, and you pick them the same way.

What you bring	How it runs	Plan
Your own key for a model maker or a model router	Runs that maker's models, or the models you name on the router, on your key. Your provider bills you, so these calls use no StellarFirm credits.	The premium makers' keys need a paid plan. The others work on every plan.
Your own endpoint, a server that speaks the common chat completions format	Add its address, a key and its model names. It shows under Your own models.	Ultra and enterprise.
A Hugging Face model	Choose Add from Hugging Face, pick a model and paste your Hugging Face token. It runs on your Hugging Face account.	Ultra and enterprise.
Your LiteLLM gateway	Choose Add from LiteLLM, give its address and key, and pick the models it serves.	Ultra and enterprise.
A model on your own computer	In the desktop app, an assistant can run a model that runs on your machine, free.	Every plan.

Auto never picks a model you brought yourself. Choose it in the picker, or rank it in an assistant's Manual list.

Hugging Face models StellarFirm runs for you#

On Ultra and enterprise plans, and for the teachers and admins of a university on StellarFirm for Universities, the picker also lists a set of open models that StellarFirm runs through Hugging Face for you, among them models for science and teaching: one that reads papers, figures and charts, and models made for Spanish, Catalan, Basque, Galician, Polish and Southeast Asian languages. Their names end in "(Hugging Face)". They use your StellarFirm credits for what each call reads and writes, like any StellarFirm model, so you need no Hugging Face account for them.

Related#

• What credits are and when they are spent: how credits work.
• Where to see what each model used: Usage.
• Reading models and datasets on the Hub: Hugging Face.