Checked again, on a schedule.
An agent that passed its review can still drift: a new permission, an edited instruction, a platform update. On each review date we re-check your agent against its Role Card and the review findings, using your own settings and activity logs, and send you a one-page report with a clear verdict. We install nothing, run nothing on your behalf, and keep no standing access.
You get: a one-page report on each review date, with every part of your Role Card marked In bounds, Changed or Out of bounds, and a verdict you can act on.
Monitor starts after an Agent Readiness Review, because that is where the baseline comes from. It is quoted from what the review finds.
Six multiple-choice questions, about two minutes. You get an answer straight away: a good fit, worth a real conversation, or not a fit right now. We only call if you ask us to.
Agents drift without anyone deciding they should.
Nothing has to go wrong for an agent to end up doing more than you agreed. It usually happens through small changes like these:
- Someone adds a connector or widens a permission to get something done.
- The platform updates its models or its agent builder.
- A tool or plugin the agent relies on moves to a new version.
- The instructions get edited.
- Approvals get switched off “just for this week” and stay off.
Your Role Card is the written record of what was agreed. Monitor checks the agent against it again.
Six checks, the same way every time.
- Access. The connectors, permissions and data sources the agent can reach, compared with the inputs on its Role Card.
- Approvals. Whether outputs marked “drafts for approval” still wait for a person, with a sample of recent actions to confirm it.
- Boundaries. Whether anything on the card’s “never” list has happened, checked against the activity log.
- Changes. What has changed in its instructions, tools and versions since the last check.
- Results. Whether the job is still getting done, using the check you wrote on the card.
- Open findings. Which findings from the review are fixed, still open, or back again.
The first four run on fixed rules. The last two need judgment, and the report says which is which.
Monthly by default, and only the access the check needs.
- How often. Monthly, on your Role Card’s review date. Quarterly for low-risk agents that only draft and never act on their own. If you tell us something has changed, like a new tool or a new permission, we can check sooner, quoted separately.
- Access. Read-only, and only during the check. Either you give us a temporary viewer role on the agent’s platform and remove it afterwards, or you export the settings and activity log using our checklist, so we never touch your systems at all.
- What we never hold. Standing credentials, your passwords, write access to anything live, or customer data beyond the sample in the activity log.
One page, and a verdict.
- A status report with each part of your Role Card marked In bounds, Changed or Out of bounds, and a one-line reason for each.
- What changed since last time: access, instructions and versions.
- Any findings, ranked by severity, each with a fix. Open findings from the review stay tracked until they are closed.
- Your updated Role Card, with the next review date on it.
- A verdict: Still within its boundaries, Changes needed, or Pause recommended. Virasai stands behind it.
- An optional walkthrough of 20 to 30 minutes, by email or video call. It is part of each cycle, and we don’t use it to sell you anything.
Nothing watches your agent between checks.
There are no alerts and none of our software in your systems. Your agent stays on your own platform and your own subscription, and we don’t run it or anything else for you.
Want a live view of where every job stands, day to day, that your own team watches? That is a dashboard we build for you and hand over.
Monitor starts with a review.
A re-check needs something to check against. The Agent Readiness Review sets that baseline, and Monitor is quoted from what the review finds.
Six multiple-choice questions, about two minutes. You get an answer straight away: a good fit, worth a real conversation, or not a fit right now. We only call if you ask us to.