AI that asks before it acts

Risotto’s agent works through the tools you choose, with the inputs you fix, after the approvals you require. Every change is on the record, and your security team can check each control in the product, not in a slide.

Book a security review Read the sandbox whitepaper

The controls, at a glance

  • SOC 2 Type II and HIPAA. Audited controls, BAAs signed on the Enterprise plan, and zero data retention with our model providers, OpenAI and Anthropic.
  • Approvals you set. A change runs only after the approvals your policy requires.
  • Permissions per department. Each team sees only its own requests, tools and credentials.
  • Verified requesters. Employees confirm who they are in your Okta before sensitive tools run.
  • Write-only credentials. No screen shows a saved secret again, and the agent never sees one.
  • A record of every change. Old and new values for every setting, and a log of what ran on every ticket.

The agent can only do what you allow

An AI agent is only as safe as the limits around it. In Risotto, IT sets those limits, and the agent cannot change them.

  • Tools you choose. Each runbook lists the tools the agent may call for that kind of request.
  • Inputs you fix. Fix the values that must not change, such as the Okta group or the Jira project. The agent never sees them as a choice. Runbooks
  • The AI can’t loosen a tool’s approvals. The Assistant can add an approval requirement to a tool, but only a person can remove one.
  • Every runbook change is a draft. You review what the Assistant wrote before it goes live, and you can restore any earlier version.

Approvals you write once and use everywhere

Write a policy once, such as “contractors need IT approval”, and attach it to access rules, tools and any runbook step. Risotto sends each approval to the right person, including the requester’s manager, in Slack, Microsoft Teams or email.

  • Block self-approval. Turn it on for an approval chain, and nobody approves their own access.
  • A checked business reason. Risotto checks the requester’s reason against your instructions for that app before it grants access.
  • Access that ends on time. Grants expire when your policy says, and a renewal can require a new approval. If an app’s removal script fails, IT gets an alert, and the grant stays open on the access grants page until someone retries it.

Sensitive tools wait for approval and proof of identity

Give any tool its own approval. Require it on every run, or attach an approval policy so that contractors need IT and employees need only their manager.

For password and MFA resets, also turn on user verification. Before the tool runs, the requester verifies in your Okta, with an Okta Verify push or a sign-in to your own Okta app. Your Okta policy decides how they prove it, and Risotto checks that the sign-in is recent and matches the requester.

Every change on the record

  • Audit log. Every configuration change in Risotto: who changed what and when, with the old and new values.
  • Audit-events API. Stream the audit log to your SIEM with a scoped API key.
  • Ticket history. Each ticket shows what the employee asked, the tools the agent ran and who approved them.
  • Access grants. See who has what, who approved it and until when. Expire access in one click, or export to CSV for your auditors.

Secrets the AI never sees

Credentials live in a write-only vault. After you save one, no screen or API shows it again.

  • Scoped. Each credential belongs to your organization or to one department, and works only with that department’s tools.
  • Locked to the hosts you list. Scripts can send a credential only to its allowed hosts and their subdomains. Every department credential must list its hosts.
  • Kept out of the AI. The agent never sees a credential. When a script tool calls an API, Risotto adds the credential only as the request leaves for an approved host, so the script never sees it either.
  • Rotated in place. Paste a new value, and everything that uses the credential picks it up on its next run. Each change is in the audit log, and stored values are encrypted.

Who can see and change what

Each department, such as IT, HR, Legal or Finance, has its own members, private queue, tools and credentials. Roles cover admins, managers and members, plus supervisors and agents in each department.

Admins and agents sign in to the dashboard with single sign-on through Okta, Microsoft Entra ID or Google Workspace, and SSO can be the only way in. Employees get help in Slack or Teams, behind your identity provider’s sign-in and MFA.

Employee data stays need-to-know

  • You choose what the agent sees. Pick the identity and HR fields it can read, such as title, department, manager, location and start date, and preview exactly what it gets.
  • Pay stays out of HR lookups. The agent’s HR lookup never returns pay, under any setting.
  • Attachments are scanned. Every file on a ticket waits for a malware scan, and a file that fails never reaches the dashboard or your ITSM.

Code your security team can read

When a step must run the same way every time, it runs as a script tool: TypeScript that IT can read, test and trace. Each run gets a fresh sandbox that reaches only the tools and hosts the script declares. A script that writes to your systems waits for approval, unless an access rule runs it to grant or remove access: then the rule’s approval applies instead.

The MCP server and the API

Manage Risotto from Claude or any MCP client. People connect with OAuth and get only the scopes their role allows, and you can revoke access at any time. API keys are scoped too, and outbound webhooks are signed. MCP server and API

Compliance

  • SOC 2 Type II. An independent auditor has tested Risotto’s controls.
  • HIPAA. Risotto signs business associate agreements (BAAs). Hazel Health, a telehealth provider, chose Risotto in part for its HIPAA support. Read the story
  • Zero data retention with model providers. Our model providers, OpenAI and Anthropic, keep no copy of your data.

Frequently asked questions

Can Risotto change our systems without an approval?

Only where your approval policy allows it. Risotto checks the policy before each change, collects the approvals it requires and records who approved. A tool marked as needing approval asks every time it runs, except when an access rule runs it: then the access rule’s approval applies, and the dashboard warns you. Access and approvals

Can the AI give itself more access?

No. The agent that answers employees cannot change runbooks, tools, permissions or approval policies. The Assistant can draft those changes. Each Assistant conversation starts read-only, so every change waits for approval unless the person using it turns on auto-accept, up to a limit set for your org. The Assistant can add an approval requirement to a tool but never remove one.

Can Risotto’s AI see our passwords and API keys?

No. Credentials are write-only in Risotto’s vault. The agent never sees them, and script tools get them only as a request leaves for a host you approved.

Is Risotto SOC 2 compliant?

Yes. Risotto is SOC 2 Type II audited.

Does Risotto sign a BAA?

Yes, on the Enterprise plan. Risotto supports HIPAA and signs business associate agreements.

Do AI model providers keep our data?

No. Risotto has zero data retention with its model providers, OpenAI and Anthropic.

Does Risotto support SSO for the dashboard?

Yes. Admins and agents sign in with single sign-on through Okta, Microsoft Entra ID or Google Workspace, and SSO can be the only way in.

How do we get Risotto’s audit log into our SIEM?

Stream it through the audit-events API with a scoped API key. Each event shows who changed what, with the old and new values.

Bring your security team to the demo

We’ll show approvals, the audit log and the vault on your own stack, and answer every question on your review list.

Book a security review