Skip to main content
For AI agents: a documentation index is available at https://docs.coverbase.com/llms.txt. This page is also available in markdown by appending .md to the URL.
This page is part of the User Guides collection and the Coverbase AI agent (Beta) set. It sits beside The Coverbase AI agent (Beta), What the agent can do and Controls, approvals and governance.
Coverbase AI is in Beta. Your Coverbase team turns it on for your organization.
The Coverbase AI agent is the first AI in Coverbase that carries out a multi-step job across the product and writes records as you. Every earlier AI feature either answers a question, proposes a change you apply yourself, or produces one output inside one feature for a person to review. The mistake people make most often is assuming the agent replaces the earlier features. It does not. Control evaluations, questionnaire autofill, Document Insights, intake matching, clause review and Radar summaries keep running as they do today, and the agent calls the same features when its plan needs them. What is new is a single run that moves between them.

​
At a glance

​
Ask Coverbase AI chat

The Coverbase AI button on a vendor, an assessment or a Radar signal opens a chat about that record. It answers in one reply, from the record and its related records, and can search the public web when your organization has web research turned on. On a vendor or an assessment, the chat can propose a small set of changes, such as updating the vendor’s profile, adding an owner, raising a finding, marking a control as an issue or not, or drafting a follow-up. Each proposal shows as a card, and nothing changes until you apply it. A drafted follow-up never reaches the vendor until someone sends it. The chat keeps working when Coverbase AI is on. Use the chat to ask about the record in front of you. Use Coverbase AI to get a job done.

​
AI inside a feature

Coverbase has used AI inside individual features for some time. Each one produces one kind of output, and a person reviews it in that feature. The Agent Ledger records them under these agents: These features decide nothing on their own beyond their output. A control evaluation sets an outcome a reviewer can overturn. An autofilled answer is a draft. The agent policies under What Agents May Do on the Agent Ledger page govern the two that can act on their own today. See Choose what agents may do. Coverbase AI calls these features rather than replacing them. When it launches an assessment, the assessment’s controls are evaluated as they would be for an assessment you launched. When it starts autofill on a questionnaire submission, autofill drafts the answers as it does when you start it.

​
Intake Agent

The Intake Agent is for requesters. It reads a request in Slack, Microsoft Teams, email or an AI assistant, fills in what the message says, asks for what is missing, and opens an intake request that the Front Door triages. It acts for the requester and stops at the intake request. Coverbase AI is for your risk team. It can open an intake request too, through the same intake portal, and then carry on: screening, the inherent risk questionnaire, submitting the request (a checkpoint), the inherent risk review, the assessment and monitoring. On a submitted request it can read the Front Door review, send the request back to the requester with questions, or deny it with a reason. Each of those always asks first. In Slack and Microsoft Teams, a direct message that starts with Coverbase AI: starts a Coverbase AI run instead of an intake request, when your organization has Coverbase AI and your role can run it. Every other message stays a request to the Intake Agent.

​
Your AI assistant through MCP

The Coverbase MCP server lets your own AI assistant read and change Coverbase records with your permissions. Every write it sends needs your confirmation first. It can also start, answer, steer and stop Coverbase AI runs (mutate(entity="agent_run", ...)). A run started from MCP is an ordinary background run with the same checkpoints, policies and ledger rows as one started in the app. See the MCP tool reference.

​
What is the same

  • Models and hosting. Coverbase AI and the earlier features run on Anthropic Claude models through AWS Bedrock in Coverbase’s AWS account. See AI governance.
  • No training on your data. Customer data sent to the models is not used to train them.
  • Redaction. When Redact personal data before model calls is on, it applies to Coverbase AI’s model calls as it does to every other model call.
  • Tenant isolation. All of it runs inside your organization’s boundary.

​
Troubleshooting

The Coverbase AI agent (Beta)

What the agent is and how a run works.

Controls, approvals and governance

How the agent’s changes are controlled and recorded.

Agent Ledger

Every action an agent took, and who decided.