GRC Agent is your AI teammate — an agentic automation that executes GRC workflow steps on behalf of your users, so your team can focus on higher-value work.
GRC Agents can be run on demand from a record or automatically from a Job. Builders can manage agents in Agent Studio.
LogicGate GRC Agents is a paid add-on. Contact your account team to activate and configure agents in your environment.
Understanding GRC Agents in Your Environment
Note: To add new agents or adjust existing configurations, contact your account team.
- Check activation status: Go to Admin > AI Control Center to see whether GRC Agents are active and which LLM model the agents run on.
-
View and manage configured agents. Go to Build > GRC Agents to open Agent Studio and see every agent in your organization.
-
How agents are set up: During implementation, your LogicGate Forward Deployed Engineer (FDE) will work with you to define:
- Trigger application and workflow step
- Agent name and instructions with necessary context
-
Post implementation support: Contact your LogicGate account team/Support if you need help create, delete or troubleshoot any agents. Builders can edit an agent including: update its instructions, change its trigger, turn it on or off, and view the skills agents can use.
-
Agent users: Each GRC agent is associated with a system-generated agent user that can be tracked and governed like a human user. To control its access:
- Add or remove it from a Permission Set to set its level of access
- Grant the agent user Edit permission on the trigger step so it can populate fields and perform link/create actions — and Edit permission on any linked workflow step if you want the agent to populate records it links or creates there. Missing the linked-step permission is the most common reason a customer sees "record created but not populated."
-
Use User Groups to restrict access to certain human users only to protect your sensitive records
See the Risk Cloud Permissions and Access help article to learn more
Managing Agents in Agent Studio
In Build > GRC Agents, click Edit Option to open Agent Studio for each agent. Builders can:
- Edit instructions. Describe the agent’s role, what it should do (e.g., intake or risk assessment, how to handle linking or creating records). Write them as you would brief a new teammate.
- Edit manual trigger. Choose the application, workflow and step where users run the agent. This stays editable until the agent is tied to a Job.
- Turn an agent on or off. An agent that is off does not run. Turning it off stops any run in progress.
- View available skills. See what the agent can do, such as filling in fields, linking or creating records, and reading reports, linked records and public web pages.
- Note: If an application is deleted, its agents are deleted. If a workflow or step is deleted, its agents are turned off and their configuration is kept.
Running a GRC Agent
You can run an agent on demand from a record at its trigger step, or automatically from a Job.
Run Agent On Demand (Manually)
-
When an agent is enabled with the manual trigger on a workflow step, the record assignee will see a Run GRC Agent banner.
-
Click Run GRC Agent. The record is temporarily assigned to the agent while it works.
-
Once execution completes, the record is reassigned to you with a status summary. The summary banner and the comment list what the agent did, the items flagged for your review, and any errors.
-
Each populated field shows a confidence score and reasoning. Sources distinguish the instructions, the current record, linked records, a table report, a relationship report, and named attachments.
-
The agent also posts a comment with the same summary, so the human assignee (reviewer) can refer back to it at any time.
-
Note: If you refresh the page, it resets to the initial run state — refer to the record comment for a persistent summary.
-
Note: If you refresh the page, it resets to the initial run state — refer to the record comment for a persistent summary.
Run Agents Automatically with Jobs
Use the Run Agent operation in a Job to run an agent automatically when a record moves or is created. There are 2 ways to add Agents to Job.
- Edit an existing Agent that has been set up to trigger manually. Navigate to Trigger section in Agent Studio. Click Job to the Job Builder Page, or click existing Job Trigger to make any edit of the current job.
- Alternatively, go to Build > Jobs and create or edit a Job.
- Choose Record Move or Record Create as the trigger. You will see "AI Available" label for Job Triggers where Run Agent Operation is available.
- Add the Run Agent operation and select the agent.
-
Note: Choose an agent whose trigger step (you set up in Manual Trigger) matches the destination step of Job Trigger, so it runs in the right use case.
-
Note: Choose an agent whose trigger step (you set up in Manual Trigger) matches the destination step of Job Trigger, so it runs in the right use case.
- Add other operations or conditions as you normally would. You can chain agents with conditions and other operations.
- Save the Job.
What Agents Can Do
Agents perform actions based on the agent prompt and the step permissions configured for your workflow.
- Autofill current record: Populates blank and supported fields on the current step of the triggering record including conditional fields the agent unhides during execution. Agent autofill will using context from: Field values, attachments, and linked workflow tables.
- Autofill child record: Populates blank and supported fields of the child record(s) agent created, including conditional fields the agent unhides during execution. Agent autofill will using context from: Field values from the parent record where the agent is triggered.
-
Link records: Links records with a strong/possible match strength (≥60%) using Record Linking Recommendations (RLR). Linking works even if RLR isn’t configured. Behavior depends on your workflow mapping between triggering WF and linked WF:
- 1-to-1 or many-to-1 mapping — links the single strongest match
- 1-to-many or many-to-many mapping — links all strong matches
-
Create records: Creates new records based on linked workflow configurations and agent prompt instructions:
- 1-to-1 or many-to-1 mapping — creates one record
- 1-to-many or many-to-many mapping — creates multiple records
-
Notes:
- Records the agent creates (e.g., a linked assessment or finding record) are left unassigned by default, rather than assigned to the agent user — this is intentional, for security.
- If both linking and creation are enabled, the agent first attempts to link strong/possible matches (≥60%). If no strong/possible match is found, it creates a new record instead
- Read reports and linked records. The agent can read table reports and relationship reports on the record, and details and attachments from linked records.
-
Read a public web page. Give URL(s) in text field(s) or in the agent’s instructions, such as a vendor trust portal, subprocessor list or policy page. The agent reads that one page.
- Note: Pages behind a login, PDFs, images and Office files are not supported. Very long pages are truncated, and the agent is told.
-
Assign and reassign records: When triggered, the record is temporarily reassigned to the agent while it works. Once execution completes, the record is automatically reassigned back to the original human assignee.
- Note: Records the agent creates are left unassigned by default, rather than assigned to the agent user.
- Post execution summary & notify reviewer: After execution, the agent leaves a comment on the record tagging and notifying the original assignee with a summary of what it did, what needs review, and any issues encountered — so the summary is accessible at any time, not just immediately after execution.
- Audit trial: All field and record changes made by the agent are tracked in Record History and Field History — the same audit trail used for human user actions — so you always have a clear record of what the agent did and when.
FAQs
1. Do I need to pre-configure Autofill or Record Linking Recommendations for agents to use those skills?
No. All agents have Autofill, Link Record, and Create Record skills built in. The agent uses them based on the prompt and step permissions — no separate configuration needed.
To get the best results, we recommend:
- Writing clear field names, guidance text, and tooltips so the agent has more context
- Configuring RLR by selecting the top 5 most descriptive fields for each linked workflow.
2. What step permissions does the agent need?
Agents follow the same step permission model as human users. Grant the agent user Edit permission on the trigger step so it can populate fields and perform link or create actions.
Grant the agent user Edit permission on the trigger step so it can populate fields and perform link or create actions on that record — and Edit permission on any linked workflow step if the agent should also populate records it links or creates there.
3. Which field types can agents populate?
GRC Agents use Autofill to populate fields, so the same field support applies:
| Support | Field Types |
| Directly supported | Text, Text Area, Radio, Checkbox, Select, Multi-Select, Number, Date Picker |
| Indirectly supported | Standard & date calculations, text concatenations, linked record counts |
| Not supported | User, Attachment, E-signature, Due Date |
Agents populate blank, visible, and editable fields on the current workflow step. When an agent unhides a conditional field during execution, it will also autofill those newly revealed fields — but only if confidence is high.
4. Does the agent only read fields that are currently visible on the step?
The agent reads the fields visible on the record’s current step, including global fields and fields in inactive sections. Fields hidden by conditions, or on other steps, aren’t read.
5. How do I troubleshoot an agent?
Go to Build > GRC Agents, click the gear icon on the top right corner to review agent status and execution details
6. What happens to an agent when its application, workflow or step is deleted?
If an application is deleted, its agents are deleted. If a workflow or step is deleted, its agents are turned off and their configuration is kept.