Policies
A policy is a rule that applies to every employee in its scope, whatever its autonomy level. Use policies for the things that must never happen, or never happen without a human, such as paying an unknown merchant or emailing a competitor’s domain.
Policy types
Section titled “Policy types”| Type | What it does | Example |
|---|---|---|
| Block | The action never runs. The employee is told why and looks for another way, or escalates | Never delete records in Salesforce |
| Require approval | The action pauses until a human approves it | Any email to more than 20 recipients |
| Limit | The action runs only within a number, a time window or a list | No more than 200 USDC per day per employee |
Scopes
Section titled “Scopes”A policy applies to a workspace, a team or a single employee.
| Scope | Who can create it | Applies to |
|---|---|---|
| Workspace | Admin | Every employee in the workspace, including marketplace hires |
| Team | Admin, or a Manager of that team | Every employee in the team and its sub-teams |
| Employee | Admin, or the employee’s human manager | One employee |
Precedence
Section titled “Precedence”Orqane evaluates every action against all policies in scope, then against the employee’s autonomy level.
- Block wins over everything. If any policy in any scope blocks the action, it doesn’t run.
- Require approval wins over the autonomy level. An employee at L3 or L4 still asks.
- Limit: the strictest limit applies. A team limit of
100USDC and a workspace limit of200USDC means100USDC. - A lower scope can make a rule stricter, never looser. A team can’t allow what the workspace blocks.
- If no policy applies, the autonomy level and its thresholds decide.
Examples
Section titled “Examples”| Policy | Type | Scope | Condition |
|---|---|---|---|
| No payments to unknown merchants | Block | Workspace | Action is Spend and merchant is not in the allowlist |
| Large refunds need a human | Require approval | Team Support |
Tool is stripe.refund and amount > 200 EUR |
| No external email outside business hours | Block | Workspace | Action is Communicate, channel is email, time outside 08:00–19:00 |
| Cap marketplace hires | Limit | Workspace | Delegate spend ≤ 500 USDC per month per employee |
| Legal review for contracts | Require approval | Workspace | Output contains a document with the label contract |
| No personal data to hired agents | Block | Workspace | Delegate step and shared data contains PII |
| Protect production | Block | Team Engineering |
Tool is github.merge on branch main |
| Daily CRM changes | Limit | Employee Leo |
Write internal ≤ 300 records per day |
Create a policy
Section titled “Create a policy”Open policies
Go to Settings → Policies and click New policy. For a team or an employee, open its page and go to Policies.
Choose the type
Pick Block, Require approval or Limit.
Define the condition
Build the condition from:
| Field | Values |
|---|---|
| Action | An impact class, such as Spend or Communicate, or a specific tool |
| Target | Domains, merchants, connectors, records, channels |
| Amount | Money, recipients, records, runs |
| Time | Hours, days, per task, per day, per month |
| Content | Labels, PII types, keywords |
Choose approvers
For Require approval, choose who approves: the human manager, a specific member, or any member with a role. Add a second approver for high-risk actions.
Test in simulation
Click Simulate before you turn the policy on. See below.
Activate
Click Activate. The policy applies to the next step of every run in scope, including runs already in progress.
Test a policy in simulation
Section titled “Test a policy in simulation”Simulate replays the last 30 days of runs in scope against the new policy, without changing anything.
You get:
- The number of steps the policy would have blocked, sent for approval or limited.
- The list of affected runs, with a link to each step in Activity.
- The estimated number of approvals per week it will create for each approver.
Use it to catch a policy that is too broad before it floods your approvers or stops production work. You can also run a policy in Monitor only mode: it records what it would have done, for as long as you need, without enforcing.
When a policy triggers
Section titled “When a policy triggers”- Block: the step is marked Blocked by policy, with the policy name. Repeated attempts on the same action within a run count as a policy violation attempt and demote the employee.
- Require approval: an approval is created and the run pauses at a checkpoint.
- Limit: the step runs if within the limit. Above it, the employee asks for approval or stops, depending on the policy.
Every trigger, change and simulation is recorded in the audit log.