For IT teams · free · Claude Code, Codex, Cursor · Claude and ChatGPT (steps not yet checked on every plan)

Your people use AI agents. Set the rules once. See who is connected and what their agents report.

agentgov gives IT one place for the company's AI rules. Switch on best-practice rules, invite everyone with one link, and follow what their agents do: who, in which app, which tools, which rule was hit. Without reading anyone's messages.

Start free See the rule sets

Rules plus watching, not a lock. No card, nothing to install for your people.

How it works

Four jobs, one page

IT signs in, creates the company and does the rest from one dashboard. Your people only open a link and follow a few steps in the app they already use.

Rules

Switch on 22 best-practice rules or write your own. Every agent reads them at the start of each session.

Invite

One link per person. They pick their app, follow a few steps and see "Connected".

Watch

Who is not connected yet, who is doing what right now, which rule was hit.

Audit

Every session, tool, request and decision, filterable and ready to export.

Rule sets

Best-practice rules, ready to switch on

Each rule says what the agent must do and, for each app, how far it goes: enforced by the app itself, partly enforced, or told to the agent and reported.

Company rules always on

Keep the agentgov connector on and read the company rules (team_rules) at the start of every session.

Enforced by Claude Code
Set in the admin console of Cursor, Claude and ChatGPT
Told to the agent in Codex and other apps

Don't open secret files

Never open .env files, private keys or credential stores, even when asked to "look at the config".

Partly enforced by Claude Code: common forms stoppedA shell command can still read the file (for example cat .env).
Told to the agent in Codex, Cursor and other apps

Destructive actions need a yes

Before deleting data, force-pushing, dropping tables or changing many records at once, stop and get an explicit yes from the user.

Partly enforced by Claude Code: common forms stoppedOther ways of deleting (rm -fr, find -delete, a script) are not stopped.
Partly enforced by Codex: common forms stoppedCodex asks before commands outside its sandbox, not before every destructive one.
Told to the agent in Cursor, Claude, ChatGPT and other apps

No customer personal data outside

Never send customer personal data (names with contact details, ID numbers, health or payment data) to a tool or service that is not on the company list. Mask it, or ask IT first.

Told to the agent in Claude Code, Codex, Cursor, Claude, ChatGPT and other apps

New automations need approval

Before building a bot, an agent, a scheduled job or an integration, ask IT with team_request.

Told to the agent in Claude Code, Codex, Cursor, Claude, ChatGPT and other apps

Report what the agent does

Call team_activity_report when work starts, before and after any step a rule marks "ask", and when it ends, with one plain sentence about the kind of work.

Reported by the app in Claude Code, Codex and Cursor
Told to the agent in Claude, ChatGPT and other apps

Where a badge names Codex, Cursor, Claude or ChatGPT it follows that app's own documentation: steps not yet checked on every plan. Claude Code is tested.

All 22 rule sets

An example

Matej and Nina

Matej runs IT. Nina works in marketing and uses Claude. This is their first week with agentgov.

Matej · IT
He creates the company
He signs in with his work email, creates the company and switches on the "Recommended start" rules. One click, one saved version.
Matej · IT
He invites Nina
He types her address. Nina gets one mail with a link, which says plainly what IT will see and what it never sees.
Nina · marketing
She connects Claude
She opens the link, picks Claude, follows a few steps in its settings and sees "Connected". Nothing to install.
Nina · marketing
She has an idea
Her agent knows that work with customer data needs IT's approval, so it asks Matej first and waits.
I have an idea: a campaign from our customer list.
Matej · IT
He says yes, with a condition
He approves with the condition "no customer data outside the company". Nina's agent follows it, and the idea can go to dataready for a test on a synthetic community before anyone builds it.
Approved: no customer data outside the company.
The dashboard
Matej sees it all
Who is connected, what their agents report, the request and his decision. On Monday the digest tells him who is still not connected.

Claude's connector steps come from its documentation: steps not yet checked on every plan.

The dashboard

One page for who is connected and what their agents report

Every row says where it came from, so you know how far to trust it: from the app, reported by a computer IT set up, or written by the agent.

agentgovYour company

Overview
People
Live
Activity
Rules
Requests
Connect

12Connected now

3Not connected

1Requests waiting

4Rule hits today2 reported by computers, not confirmed

Person	App	What	Source	When
Nina	Claude	drafting an email to a client	Written by the agent	now
Matej	Claude Code	`Bash` ×6	Reported by a computer · Laptops	1 min
Luka	Cursor	Rule hit: Only approved connectors	Reported by a computer, not confirmed	4 min

A picture of the dashboard, with example people. Claude and Cursor rows: steps not yet checked on every plan.

Overview
People
Live
Activity
Rules
Requests
Connect

A tour of the dashboard

Apps

Works where your people already work

agentgov is a connector your people add to the AI app they already use. Some apps let IT go further with settings files; we say exactly how far, app by app.

ClaudeRules told to the agent; it reports what it does through agentgov. (steps not yet checked on every plan)
ChatGPTRules told to the agent; it reports what it does through agentgov. (steps not yet checked on every plan)
Claude Code testedSome rules enforced by the app when IT deploys our settings file, others partly; every tool reported by name.
CodexSome rules enforced by the app through IT's requirements file; tools reported by name. (steps not yet checked on every plan)
CursorRules told to the agent; sessions, shell runs and connector calls reported by name through IT's hooks file. (steps not yet checked on every plan)
VS CodeRules told to the agent; it reports what it does through agentgov. (steps not yet checked on every plan)
WindsurfRules told to the agent; it reports what it does through agentgov. (steps not yet checked on every plan)
Other appRules told to the agent, as long as the app can add agentgov by its address. (steps not yet checked on every plan)

How people connect

Privacy

What IT sees, and what it never sees

The same two lists are on every invitation and join page, so each person knows before they connect.

What IT sees

That you joined, your role, and the display name IT gives you.
Which AI app you use agentgov in, and when.
The names of the tools and connectors your agent uses.
On a computer IT has set up (Claude Code, Codex, Cursor): every tool your app uses, by name, reported by the computer.
The list of servers the agentgov app on your computer reports, if you send it.
Which company rule was hit.
Your requests and IT's answers.
Your "I'm stuck" notes.
One short sentence your agent writes about the kind of work ("drafting an email to a client").

What IT never sees

Your messages.
What the AI answers.
Your files, and what you paste.
Commands and their details.
Tool inputs.
Your passwords.
agentgov does not receive or store them.

Everything is stored in the EU. Activity is kept 180 days at most, and your organisation can choose less. Reporting from computers set up for Codex and Cursor follows their documentation: steps not yet checked on every plan. Details on the privacy page.

The limit, plainly. Rules plus watching, not a lock. agentgov tells every agent the company rules and shows IT what happened. Where IT deploys our settings files, the app itself enforces some rules. A person can still open another chat, another app or another computer.

Free

Free, for IT teams

agentgov for IT is free: the rules, the invitations, the dashboard, the activity log and the settings files. No card.

Start free

Questions

Good questions, short answers

Can agentgov stop a person?

Not by force. Every agent with agentgov reads the company rules and asks before work that needs approval. Where IT deploys our settings files in Claude Code, Codex or Cursor, the app itself enforces some rules and partly enforces others. A person can still open another chat, another app or another computer: rules plus watching, not a lock.

What do you store about my people?

Their email, role and the display name you give them; which AI app they use agentgov in, and when; the names of the tools and connectors their agent uses; rule hits; their requests and your decisions; their "I'm stuck" notes; and one short sentence their agent writes about the kind of work, cleaned of addresses, numbers and links. It is stored in the EU. Details on the privacy page.

Do you read messages or files?

No. agentgov never receives prompts, answers, files, commands or tool inputs. On computers IT sets up, a small reporting script reads the app's report on the computer, keeps only the names (the session, the tool, the connector, the person) and sends nothing else; our server refuses anything else.

Which apps work, and how well?

Claude Code works best and is tested: with IT's settings file the app enforces some rules itself and reports every tool by name. Codex and Cursor report sessions and tools through IT's settings and hooks files. In Claude, ChatGPT, VS Code and Windsurf the rules are told to the agent and it reports what it does through agentgov. For every app but Claude Code: steps not yet checked on every plan. See Connect.

Do my people need to install anything?

No. Each person opens their invitation, picks the app they use and adds agentgov in its settings in a few steps. The settings files for Claude Code, Codex and Cursor are optional; IT deploys them the way it deploys other settings.

What does it cost?

Nothing. agentgov for IT is free: the rules, the invitations, the dashboard, the activity log and the settings files. No card.

How long do you keep the activity?

180 days at most; your organisation can choose 90 or 30. Requests, decisions and the admin log are kept 3 years, "I'm stuck" notes 90 days. Deleting the organisation deletes everything at once.

Is this legal advice for monitoring at work?

No. Tell your people what you watch; the join page tells them too. Check your local rules with your lawyer.

Built with prodready? Tested with dataready? Governed with agentgov.

prodready makes what you built safe to put in front of real users. dataready tests an idea on a synthetic community before you build it. agentgov keeps your company's AI agents to your rules. Made by the same team.

Visit prodready.si