Feature Request: Provider-Level Restrictions and Default Model Settings for Workspaces | Dust Community

Daphné
·
·
HI, you can already do the first one from the Model Providers page 🙏🏻
Damien Floch
·
·
Hi there, I've sent an email over as well but just to second the need on being able to set a default model for the "Standard" and "Premium" tiers especially, for consumption control and risk management (to right size the models behind these tiers to our workspace needs/put smaller models where applicable). Thanks!
Daphné
·
·
Thanks for the additional context. Setting a custom default model behind Standard and Premium isn't planned right now, but we've logged the request. 🙏🏻 Each of these auto models is an ordered list of candidates that Dust maintains and publishes in the docs, and it resolves to the first model enabled in your workspace. So while you can't pick the model directly, disabling providers on the Model Providers page does change what Standard and Premium resolve to. For agents you build centrally, you can also pin a specific model. Out of curiosity, if you could set them today, which model would you put behind Standard and Premium, and why?
🙏1
Damien Floch
·
(edited)·
Thanks Daphné, from our side specifically, GPT 5.6 Luna High for Standard strikes the right balance between performance and consumption. I'm more worried about future evolutions and if you decide to e.g. to put a bigger model/put back Claude Sonnet, which would have a huge impact on consumption on our side. In that case, I'd like to retain a Luna tier model for Standard. For Premium, I ended up disabling it entirely, to ensure only specific agents/use cases are running these higher-tier models (rather than having broad access for every agent, which was creating a lot of unpredictability in consumption). On the specific Premium model, perhaps either GPT 5.6 Sol* (6 now seems more to belong to the Standard tier, and is actually cheaper than Terra) or the latest Opus, but just with Low reasoning instead of defaulting to High. The underlying fallback mechanism is great to have. Could the feature be about letting admins decide which specific models this mechanism resolves to first? Having that kind of setting would put me much more at ease regarding consumption control and any future evolutions you make of the tiers; especially Standard ; with the overall goal being to let admins more broadly define the model in these tiers according to their workspace's own needs. Happy to provide more details if helpful!