Models
Sail doesn’t hold any model keys. Every model comes from your Msty Nexus: whatever Nexus offers, Sail can use. Adding a model, or a whole new provider, happens in Nexus.
What you decide in Sail is who uses which model, and for what.
See what’s available
Section titled “See what’s available”Go to Admin → Nexus. Under Models, you see every model Sail can use: its name, ID, format and how much text it can take in (Context).
A model marked Not usable by Sail speaks a format Sail doesn’t support yet.
Added a model in Nexus? Click Check again to see it here.
The three places models are chosen
Section titled “The three places models are chosen”| Where | What it decides | Who sets it |
|---|---|---|
| The default assistant | The model most chats use, since every new chat starts with it | Owners and admins |
| Each assistant’s model list | Which models that assistant can use, and which one is its default | Whoever maintains the assistant |
| Utility model | The background model that writes titles, saves memories and summarizes long chats | Owners and admins |
The default assistant
Section titled “The default assistant”Most of your company’s chats run on it, so give it a strong all-round model.
To change it, go to Assistants, open Sail, click Edit, and change its Models. Publish, and every chat with Sail uses the new model from its next message.
Each assistant’s models
Section titled “Each assistant’s models”Every assistant has an ordered list of up to eight models. The first one is its default. See Assistants → Which model should answer? for labels, hints and reasoning effort.
Match the model to the job, not the other way round:
| Job | A good fit |
|---|---|
| Quick everyday answers, drafting emails | A fast, low-cost model |
| Contract review, forecasts, anything with numbers | A stronger model that reasons, with Reasoning effort set higher |
| Anything too sensitive to leave your network | A model running on your own hardware, through Nexus |
| Long documents | A model with a large context window |
When you give an assistant several models, label them for people, not by model name. “Balanced”, “Deep” and “Private” mean more to a finance analyst than a model ID.
The utility model
Section titled “The utility model”Go to Admin → Settings → Utility model.
Pick a small, fast model that follows instructions well. It runs after nearly every reply, so speed and cost add up. But don’t go too small: a weak utility model makes memory worse, because it decides what’s worth remembering.
Left on Same as the default assistant, Sail uses the default assistant’s model for this background work.
When a model goes away
Section titled “When a model goes away”If Nexus stops offering a model:
- Chats with an assistant move to the next model in its list, on their own.
- The assistant’s owner sees a warning on its page, naming the missing model.
- Sail never quietly switches to a model outside the list. If nothing in the list is available, people see a clear message instead.
So give important assistants a second model as a fallback.
Let people pick any model
Section titled “Let people pick any model”By default, people can only choose from an assistant’s own list. That keeps answers consistent and costs predictable.
Some people need more freedom, like a team evaluating models. For them, turn on Pick a raw model in Policies, for everyone or for chosen teams. They then see Any model in the message box, with every model Nexus offers.
Control spending
Section titled “Control spending”Sail has no per-model spending limits. Set those in Nexus, which sees every request.
In Sail, you can:
- Limit Messages per hour per person in Policies.
- Watch costs in Admin → Usage, by person, team, assistant and model. See Audit log and usage.