Permissions & Limits
This page gathers who can use each part of automation and the limits that apply to it. Features are granted through staff roles; see Features & Permissions.
Permissions
| Feature | Grants |
|---|---|
automation.automation_access |
Workflow rules, running workflows, conditions and action logs, webhooks, and the Automation tabs on forms and events. Marked internal |
core.institutions.classes_automation_change |
Setting up workflows for specific classes. Includes automation.automation_access |
core.institutions.institutions_lifecycle_workflows_change |
Settings → Automation & AI → Lifecycle workflows. Adding a workflow to a group also needs automation.automation_access |
messaging.messaging_change |
The Publishers tab of Workflow rules |
automation.agents_management |
Settings → Automation & AI → Agents: creating and editing agents, their runs and change history, and the active AI provider |
core.institutions.profile_classes_change |
Applying an agent's potential suggestion from a journey card, as for any potential change |
tasks.tasks_access, tasks.tasks_change |
Tasks. See Tasks |
The Workflow rules item in the settings menu also needs the permission to change institution settings. The Workflows tab on a profile is shown to administrators only, and system workflows and system webhook subscriptions are listed only for administrators and support staff.
Agent results (scores, verdicts, potentials) are shown to staff wherever they can see the application or person they belong to; viewing them needs no automation permission.
Workflow Limits
| Limit | Behaviour |
|---|---|
| Filter checked once | Conditions are checked when the trigger fires, never again |
| Steps copied at entry | A workflow keeps the steps it started with; later changes to the template apply only to new workflows |
| No step retries | A failed step puts the workflow in Error; the remaining steps do not run |
| Exit conditions | Checked only once the first step has started |
| Segment workflows | A segment or segment workflow changed in the last 15 minutes is not checked until it settles |
| Processing | Due workflows are picked up continuously, in batches of up to 500 |
| Scheduled triggers | Every minute, hourly at 40 minutes past, daily at 05:00, server time. Not available to staff-built workflows |
| System workflows | Cannot be suspended, resumed, duplicated or deleted |
Webhook Limits
| Limit | Behaviour |
|---|---|
| URL | http:// or https:// only; cloud metadata and link-local addresses are refused |
| One per URL | One subscription per event and URL |
| Timeout | 5 seconds to connect, 10 seconds for the answer |
| Retries | Twice more on a network error or a 5xx answer; never on a 4xx |
Agent Limits
| Limit | Behaviour |
|---|---|
| AI model access | The application scorer and potential scorer run only while AI model access is on. Turning it off disables them, and they stay disabled until an administrator enables each one again |
| One run at a time | While a run for an agent and a record is waiting, new events for that record join it |
| Attempts | A failed run is tried up to three times in all |
| Runs shown | The runs page shows the latest 200 runs of an agent; Change history the latest 200 changes |
| One result per record | A new run replaces the agent's previous result for the same record |
| Attachments | The application scorer reads the text of PDF attachments only, about 10,000 tokens per file and 50,000 in total, within 60 seconds. Password-protected or damaged files are skipped. Files themselves are never sent |
| Potential scorer input | The person's 20 most recent applications, event registrations and form submissions. Contact details and identity fields are left out |
| Staff are never scored | The potential scorer skips staff members and erased people |
| Document scans | One run per uploaded file. A scan that runs past its time limit is recorded as invalid with the reason, and the run fails |
| Erasure | Deleting or anonymising a person deletes every agent run about them, its AI audit and their scores. Deleting an application deletes the results recorded on it |
Note: Every run of an AI-capable agent, with the prompt sent and the answer received, is kept for review. See AI Console & Assistant.