AI Settings
AI settings let you manage the output of AI summaries and the default model configuration the AI agent uses. Use them when you want to tailor the wording of deal analysis to your company, or switch models on a per-skill basis.
Open the settings screen
- Open workspace settings
- Select "AI Settings"
AI summary prompt
You can change the system prompt the AI uses when generating deal summaries.
- Turn on "Use a custom prompt"
- Edit the prompt in the text area
- Click "Save"
If you want to revert to the default, click "Restore default."
What to specify in the prompt
| Item | Example |
|---|---|
| Output structure | Deal health, highlights, risks, what to do today |
| Metrics to emphasize | Viewing time, customer replies, overdue tasks, decision-maker involvement |
| Tone | Short, actionable language a sales rep can act on next |
| Handling of rationale | Separate speculation from fact and note the supporting evidence |
The AI summary is automatically provided with viewing logs, task status, message history, basic Room information, and more. Write the perspectives you want it to reference in the prompt to adjust the output.
Default AI provider
Users with administrator privileges can choose the default provider the AI agent uses.
| Option | Use case |
|---|---|
| OpenAI | The default setting using GPT-family models. Good for fast response generation and everyday analysis |
| Anthropic | A setting using Claude-family models. Good for complex analysis and organizing long text |
The provider setting is the default value for the entire workspace. If you want to use a different provider for a specific skill, override it under "Per-skill agent settings."
Default AI model
You can choose a model from three tiers depending on the use case.
| Model tier | Use case |
|---|---|
| Haiku / GPT-4.1-mini | Fast and low cost. Everyday summaries and lightweight classification |
| Sonnet / GPT-4.1 | A balance of quality and speed. Deal analysis and generating proposal rationale |
| Opus / o3 | For situations that require deep reasoning. High quality, but watch your usage |
Per-skill agent settings
You can override the provider and model on a per-skill basis, such as for signal detection, task planner, and message planner.
| Setting | Description |
|---|---|
| Provider | Choose from Default, OpenAI, or Anthropic |
| Model | Choose from Auto, Haiku/GPT-4.1-mini, Sonnet/GPT-4.1, or Opus/o3 |
| Auto | Follows the workspace default setting |
Splitting work—lightweight models for daily digests and routine summaries, high-quality models for proposal rationale and deal-progression decisions—makes it easier to balance cost and quality.
Related settings
| Setting | Role |
|---|---|
| Agent Policy | Manages autonomy level, web search, rules, and glossary |
| Custom Skills | Adds your company's own judgment logic and prompts |
| Deal Progression | Manages phase definitions, completion criteria, and concrete actions |
Regardless of model settings, output involving customer-facing messages, materials, or contract terms should be reviewed by a person before approval.