The agent works in a private sandbox for each project, and your data is never used to train models.
Scope
Each chat reads one project, chosen before the first message and fixed after it. The agent can use:- Accounts and sessions, with their scores, signals and list membership
- Rules, their definitions, and every time each one fired
- Verifications, including ID Check and Face Match photos
- Live state from the Verisoul API, such as a list’s current members
- The Verisoul documentation, for product and integration questions
@, paste up to 1,000 IDs in one batch, or attach files (up to 20 per message, 25 MB each).
Prompts worth trying:
- Pull all accounts linked to
{account_id}at above 50% strength, three levels deep - Compare fake and suspicious users over the last 7 days against the prior 7. What is driving the change?
- Here are 100 emails. Find their accounts and add them to the block list.
Outputs
The agent does not create or turn on rules. Review in Rules opens the rule builder with the suggestion filled in and backtesting running, and you save it from there.
Changing lists
The agent can create lists and add or remove accounts, from pasted IDs, attached files, or its own findings.
An approval card lists the change. Approve runs it and Reject discards it.
List changes through the agent need the Change Allow and Block lists permission (Editor and Admin). For other roles, chats are read-only. Analysts can still manage custom lists from the Lists page.
Analyze with AI
On an account, the analysis reads your project’s scoring thresholds, the lists the account is on, and the rules it hit. On a verification session, it also reads the captured photos and the session video. Open full chat continues the conversation in AI Agent. Incomplete verification sessions cannot be analyzed.Memory
The agent keeps notes between chats:- Per person: stated preferences, such as preferred chart types or recurring breakdowns
- Per project: what it has learned about the project’s data and fraud patterns, shared with everyone who uses the agent on that project
Sharing
Only the person who started a chat can share it.
Turning off Share publicly disables the link immediately. Turning it on again creates a new link.
Scheduled runs
A prompt can run on a schedule through the MCP server, usingagent_schedules_create with a cron expression in UTC or as a manual schedule. Each run opens a new chat for the schedule’s creator and uses that person’s permissions. Creating schedules needs Schedule agent runs (Editor and Admin).
Permissions
Admins have every permission. See Managing Users for custom roles.