Models & settings
Use the model picker for each agent request, then use Settings for defaults that shape the workspace.
Choose a model#
Open the model picker in the agent composer. It lists the managed models currently available to your account; that list is the source of truth because availability can change.
Choose based on the work, not the brand name:
| Work | Practical starting point |
|---|---|
| Direct question or small rewrite | Faster model, low reasoning |
| Multi-file comparison | General model, medium reasoning |
| Complex Office edit or research synthesis | Stronger model, medium or high reasoning |
| Repetitive cleanup with clear instructions | Faster model, low reasoning |
If a result is weak, first improve the context and constraints. A larger model cannot fix an ambiguous source of truth.
Set reasoning effort#
Reasoning controls how much internal work the model can spend before responding. Use Low for routine edits, Medium for cross-file work, and High only when the task genuinely benefits from deeper analysis.
High reasoning can take longer and use more of the account's allowance. For a simple format change, it can also produce more process than the task needs.
Configure the workspace#
Open Settings with Cmd/Ctrl + ,.
Settings is organized into these sections:
- Appearance — theme and interface presentation.
- Editor — source editor behavior such as type size, line height, and wrapping.
- AI — agent defaults, including reasoning and planning behavior.
- Files — file and project preferences.
- Shortcuts — searchable keyboard reference.
- Account — sign-in, plan, and usage information.
- About — app version and product information.
- Help — feature guidance and support routes.
Planning mode#
Enable planning in Settings → AI when you want the agent to propose steps before it acts. Planning is useful for multi-file deliverables and risky restructuring; it is unnecessary for most direct edits.
Ask explicitly when approval matters:
Plan the workbook cleanup first. Identify formulas, named ranges, and charts that could be affected. Wait for approval before changing any cell.
Accounts and model access#
Nebula Writer routes agent requests through its managed service. It does not currently offer settings for personal OpenAI, Anthropic, or other provider API keys.
The Account section shows the plan and current usage state. Model access, request limits, and collaboration entitlements can differ by plan, so rely on the controls and messages in the current app rather than a fixed list in a saved guide.
Reset a confusing session#
When a conversation is carrying the wrong assumptions:
- Start a new chat.
- Open the authoritative source file.
- Add only the attachments or
@mentionsneeded for the task. - State what the agent must not edit.
- Use a low-risk first request, such as an outline or discrepancy list.
Changing models mid-conversation does not remove the existing conversation context.