Agent Policy
Agent policy lets you manage, at the workspace level, how autonomously the AI agent can operate, which skills it can use, and whether it may use external web search.
Open the settings screen
- Open workspace settings
- Select "Agent Policy"
- Change the items you need and click "Save"
Only workspace administrators can change these settings.
Agent autonomy level
| Level | Name | Behavior |
|---|---|---|
| 0 | Manual approval only | Every action requires human approval before it runs |
| 1 | Auto-run low-risk | Only low-risk internal processes run automatically |
| 2 | Auto-run medium-risk | Actions up to medium risk run automatically |
| 3 | Fully autonomous | All actions run autonomously |
For early rollout, we recommend level 0 or 1. Actions such as creating tasks, sending customer-facing messages, publishing materials, and updating stakeholders affect customers and Room content, so it's safer to start with an approval-based workflow.
Enabled skills
Choose which skills the AI agent can use. You can enable up to 20.
| Skill | Role |
|---|---|
| Deal progression | Suggests the next action based on phase, completion criteria, customer requests, and sales commitments |
| Signal detection | Detects changes in customer behavior and risks |
| Pull dialogue | Interprets the user's question intent and switches between answering, analyzing, and suggesting |
| Stakeholder analysis | Organizes contacts, roles, influence, and unidentified people |
| Task planner | Suggests required tasks with priorities |
| Pipeline digest | Summarizes key points and risks across multiple Rooms |
| Message planner | Drafts customer-facing messages |
| Content creation | Assists with creating Pages, materials, meeting notes, and proposals |
Web search
When the information in a Room alone isn't enough to answer or analyze, you can allow the AI agent to use external web search.
| Setting | Description |
|---|---|
| Allow web search | Lets the agent use external search when it determines it's needed |
| Monthly search limit | Sets the monthly search limit for the entire workspace |
If the Brave Search API key is not configured, no searches run even when web search is allowed.
Agent rules
You can register up to 5 operational rules to factor into the agent's decisions.
Examples:
- Always turn customer-facing messages into approval cards
- For proposals involving contract terms, prioritize legal review tasks
- Don't make definitive claims when speculating about pricing or discounts
- When the decision-maker hasn't joined, prioritize stakeholder confirmation
Glossary
You can register up to 10 of your company's own terms and abbreviations. The agent references this glossary when generating answers, analyses, and the rationale behind its suggestions.
| Term | Example definition |
|---|---|
| Champion | An internal advocate within the customer's organization who drives adoption |
| PoC | The validation phase before full deployment |
| Ringi | The customer's internal approval process |
Recommended initial settings
| Item | Recommended value |
|---|---|
| Autonomy level | Level 0 or 1 |
| Web search | On only for workspaces that need it |
| Enabled skills | Deal progression, Pull dialogue, Task planner, Message planner |
| Rules | Customer-facing output requires approval; rationale visible to internal users only |
The closer you get to full autonomy, the more convenient it becomes, but the impact of an erroneous action also grows. Start with an approval-based workflow for customer-facing messages, material publishing, contract terms, and CRM updates.