Models: defaults, custom endpoints and local models
Choose the default model for your Nebula workspace, add models from a catalog or your own endpoint, run one on this computer, or pick one per task.
You can choose the model your agents and tasks use by default, add models from a catalog or your own OpenAI-compatible endpoint, and run a model on this computer. Everything lives under Settings → Workspace → Models.
If you never touch this page, work still runs well. Nebula's own models are built in, and the workspace default covers anything nobody has pinned.
Nebula Flash and Nebula Ultra
Nebula's own models sit at the top of the list with a Built-in badge. There's nothing to add or enable.
The starting default. Fast, and the right choice for everyday work.
Goes deepest. Use it for hard problems where a better answer is worth a longer wait.
The workspace default
A star marks the default: the model an agent or task uses when nobody has pinned one. Changing it moves every unpinned agent and task in the workspace, so it's a workspace admin's call.
Everyone else can open the page and see what's available, but only admins change the catalog and the default.
Two kinds of model can't be the default, because the default has to run for everyone:
A model set to Only me is invisible to the rest of the workspace, so it can't serve their work.
A model locked to one device only runs there. Work anywhere else would fail, so it can't be the default.
Add a model
Choose Add model. The dialog has two tabs.
Search models from OpenRouter, Vercel and other providers, select the ones your workspace wants, and choose Save models. They show in the list with a Catalog badge.
Point Nebula at any OpenAI-compatible endpoint you control — a self-hosted model, a provider that isn't listed, or your team's own gateway.
Add a custom endpoint
Give it a Display name you'll recognise in model menus, such as "Northwind Llama box".
Enter the Base URL, such as https://api.example.com/v1, and the Model id your endpoint expects.
Add Auth headers if the endpoint needs them. The field starts with an Authorization header ready to fill in.
Optionally restrict it to one device. When you do, only that device runs the model, and work from anywhere else fails rather than quietly using a different model.
Under Who can use it, choose Everyone in this workspace or Only me, then choose Add model.
An endpoint on localhost means that device's own machine, so a loopback address needs a device. Models on local devices are always private, because only the device's owner can reach them.
Run a model on this computer
Choose Run models on this computer to run a model on your own hardware, with no provider account. Nebula installs and manages Ollama for you.
Nebula checks this computer and lists models to download, best first. The one marked Best for this computer fits your graphics memory.
Choose Download. Nebula downloads the model and prepares it so agents can use tools with it.
When setup says Ready, the model appears in the list with a Local badge.
Pick something bigger than your computer can hold and Nebula warns you that it may run slowly or fail to load. Nebula manages this model for you, so you can't edit or remove it like a custom endpoint.
Pick a model for one task
You don't have to change the default to try a different model. In a task's composer, open the model menu and choose All models… to search every model your workspace offers.
Choose This one to use the model for this task only. Workspace default saves it as the default for everything with nothing pinned. Use default returns the task to the workspace default.
Media models
Below language models, Media models sets the default for each kind of media work. Admins choose which provider models are available with Edit list and pick a default for each kind. Everyone else can see what's active.
| Kind | Used when |
|---|---|
| Image generation | An agent generates new images |
| Image editing | An agent edits or composites images |
| Video generation | An agent produces video clips |
| Video analysis | An agent analyzes or describes videos |
| Music generation | An agent composes music |
| Sound effects | An agent creates sound effects |
| Text to speech | Spoken output needs a voice |
| Transcription | Audio needs turning into text |
| Voice isolation | Speech needs separating from background noise |
FAQ
Not sure which models your workspace offers? Ask Nebula. For anything else, email support@nebula.gg.
Related
Skills and slash commands
Write a skill once and run it as a /command — on one channel for everyone, just for you, or built into an agent — and see which skill wins a name.
Notifications: destinations, rules and write approval
Get Nebula notifications on your phone or in Slack, Discord or Telegram DMs, choose which events reach you, and turn on write approval for risky actions.