Skip to main content
Use AI Governance from your AI

Use AI Governance from your AI

People who hold an AI Governance role can use AI Governance from their own AI tool, such as Claude, Copilot or Cursor. They connect the AI tool to StackJack and ask about findings, usage, spend,…

Written By Christopher Scaminaci

Last updated 4 days ago

People who hold an AI Governance role can use AI Governance from their own AI tool, such as Claude, Copilot or Cursor. They connect the AI tool to StackJack and ask about findings, usage, spend, approvals, incidents, alerts and readiness. They can also make the everyday decisions, without opening the Portal. This page covers how to connect, the access setting, the built-in connector, how compact catalog mode reaches the tools, the tools by group and what the AI tool says when it cannot help.

The tools are the built-in AI Governance connector. It needs no key and nothing to buy.

At a glance

  • What it does: lets a signed-in person read and act on AI Governance from an AI tool.
  • Plan: every plan. A few tools need Business, as in the Portal.
  • Who: a person who holds an AI Governance role and signs in with a second factor. An Administrator sets the access mode. An Auditor can read it.
  • Where you set it up: AI Governance → Settings → AI assistant.
  • Needs: StackJack added as a connector in the AI tool. The AI Governance connector inside StackJack is built in: nothing to set up and nothing to buy.

Connect your AI tool

Add StackJack as a connector in the AI tool and sign in when it asks. Your MCP endpoint and the guide for your AI tool say how. Sign in the way you sign in to AI Governance: through your organization's single sign-on, or with a second factor. Automations and shared endpoint keys cannot use these tools, because a governance decision is a person's decision. A second-factor sign-in counts for 30 days by default, and a connection that refreshes keeps its original sign-in, so a refresh does not renew it. When the sign-in is too old, the AI tool shows StackJack's refusal as the answer to your request. It does not start a new sign-in for you. To fix it, disconnect StackJack in the AI tool, connect it again, and sign in with single sign-on or a second factor.

Your organization's setting

Open Settings → AI assistant. Choose how AI tools may use AI Governance:

  • Read and act. AI tools can read AI Governance data a person's role allows, and make the everyday decisions: decide a tool request, assign or close findings, open an incident, ask to see a name and acknowledge or resolve an alert.
  • Read only. AI tools can read AI Governance data a person's role allows. Decisions, such as approving a tool request or closing a finding, stay in the Portal.
  • Off. No AI tool is offered anything from AI Governance. Your governance data stays in the Portal.

Nobody has changed this setting until an administrator does, so the default applies: read and act.

Why the setting exists. An AI tool belongs to a third party. If your organization wants AI Governance data to stay in the Portal, for example by agreement with your works council or your data protection officer, switch it off here. The setting is the same on every plan.

Who can change it. Only an AI Governance administrator can change this setting. Every change is recorded in your audit log with the old and the new choice. An auditor can read it. An AI tool can never raise its own access.

The tools that exist today

The tools are one built-in connector, AI Governance. Every tool name starts with aigov_. There are 110 tools: 60 read and 50 change something. The role column says who can use each one. "Any role" means any AI Governance role.

The connector needs no key, no setup and no purchase. It is there whenever AI Governance is available to your organization, on every plan. Its calls are never counted as tool usage and never billed. It shows on the Connectors page as one "Built in" line, with a button to this page, and never as a card you add. Nothing here gives a person more than their AI Governance role and the access setting allow. Every call is checked again when it runs.

In compact and minimal catalog mode

Some AI tools can load only a limited number of tools, so StackJack offers a smaller catalog in compact and minimal mode. See Catalog modes. In those modes your AI tool does not list the AI Governance tools one by one. It reaches them through StackJack's catalog tools:

  • stackjack_search_tools finds a tool by what you ask for, for example "list open findings".
  • stackjack_run_tool runs the tool it found.

The same checks apply as for a direct call: your role, the access setting, the second factor and the plan. If you are not admitted, the search does not find these tools at all.

Overview

ToolWhat it doesChanges somethingWho
aigov_get_summaryThe Overview, card by card: health, what needs attention, the trend, risk, AI tools, coverage, compliance, spend, the plan, Deploy and manage and the five setup milestones, with the change over 7 days. Findings caught in test mode come as a second figure beside the open ones, and the tools, deploy and setup figures come in words too. A card your role does not cover, or a feature that is off, says so.NoAny role
aigov_get_setup_statusWhere Get started stands, step by step, and what to do next. It also says how many of the sources you use work, the six places where AI use can happen and which source covers each, and the open gaps.NoAdministrator
aigov_get_planThe plan, staff against the included number, what the plan says about stored text and the plan each feature needs.NoAny role
aigov_get_healthWhether AI Governance is working, with the reasons when it needs attention. Its source counts include the switched-on sources that read well and have no event yet.NoAny role
aigov_describeWhat each area and feature does, how to turn it on, where its data shows and the plan it needs.NoAny role
aigov_get_usageCalls, tokens and findings per day, by tool, provider, model, client and team. Never a ranking of people.NoAny role
aigov_get_spendWhat AI use costs, by tool, vendor, model and client.NoAny role
aigov_get_coverageWhether monitoring covers your organization and where it does not.NoAny role. The sections follow your role.
aigov_list_organizationsThe organizations you can ask about: your own and each client organization you can see.NoAny role
aigov_create_client_organizationCreate an organization for a client, with the jurisdiction you state.YesAdministrator
aigov_open_ticketOpen one ticket in your PSA about an AI tool or a finding, through a ticket channel you chose. The ticket holds fixed sentences, no prompt text and no names.YesAdministrator. Needs the Pro plan and a second factor.
aigov_ask_for_decisionAsk the approvers to approve or deny an AI tool. Nothing is approved until they decide.YesAdministrator. Needs the Pro plan and a second factor.
aigov_get_billingThe plan, staff and the text stored with StackJack, with each client organization's usage. No payment ids.NoTenant-wide roles only. A role limited to some client organizations is refused.
aigov_get_ai_tools_modeWhether AI tools may use AI Governance. Read only: only an administrator changes it, in the Portal.NoAdministrator, Auditor

Findings and alerts

ToolWhat it doesChanges somethingWho
aigov_list_findingsSearch findings and see where each stands in review and who it is assigned to.NoReviewer, Administrator
aigov_update_findingsAssign findings, start a review, close them with a reason or open them again.YesReviewer, Administrator
aigov_list_alertsThe alerts that are waiting for someone. One alert also carries what it is about: the computer, the people, the source, the rule or the thing it names, with names as your name policy allows.NoAdministrator, Reviewer. A Reveal approver sees the alerts about name requests.
aigov_update_alertAcknowledge, resolve or reopen an alert.YesReviewer, Administrator. A Reveal approver for the alerts about name requests.
aigov_list_alert_channelsWhere alerts are sent and how healthy each channel is. Never an address or a secret.NoAny role
aigov_list_alert_policiesWhich alerts go to which channels, and how.NoAny role
aigov_get_alert_scheduleThe time zone, quiet hours and digest times for alerts.NoAny role
aigov_set_alert_scheduleChange the time zone, quiet hours and digest times.YesAdministrator
aigov_test_alert_channelSend one labelled test alert to a channel.YesAdministrator
aigov_pause_alert_channelHold the alerts of one channel.YesAdministrator

Incidents

ToolWhat it doesChanges somethingWho
aigov_list_incidentsThe incidents that are open and how they stand.NoAny role
aigov_open_incidentOpen an incident record for something that happened.YesReviewer, Administrator. Needs the Business plan.
aigov_update_incidentChange an incident's severity, title, summary, owner or notice flags. It cannot close it.YesReviewer, Administrator
aigov_add_incident_noteAdd a note to an incident's timeline.YesReviewer, Administrator

AI tools

ToolWhat it doesChanges somethingWho
aigov_list_inventoryThe AI tools and services seen in use, and the people who use them.NoAny role. People are shown to Administrators and Reviewers only.
aigov_add_inventory_itemAdd a tool to the inventory by hand, as pending.YesAdministrator
aigov_confirm_risk_classRecord the EU AI Act risk class of an inventory item.YesAdministrator
aigov_list_approvalsThe requests to use an AI tool and where each stands.NoAdministrator, Reviewer, Auditor
aigov_decide_requestApprove or deny a request to use an AI tool.YesAdministrator
aigov_propose_vendor_actionPropose removing a user, turning off a key, setting a spend limit or cancelling a seat in a vendor account, or blocking (or unblocking) a web app's domain in DNSFilter, Cloudflare or Cisco Umbrella (give the domain). An administrator approves it in the Portal; the assistant can never approve.YesAdministrator. Needs the Business plan and vendor changes turned on for the connection.

Exposure scan

ToolWhat it doesChanges somethingWho
aigov_get_exposure_scanThe latest finished exposure scan of an organization (your own, a client or a prospect), or the one you name: the exposure and visibility scores, each area with its findings and what could not be checked, the counts by risk, what each part read and up to 200 items. Never a person, a key or a scope list.NoAdministrator, Auditor
aigov_compare_exposure_scansWhat changed between the baseline scan and a later one: added, removed and changed findings, how a figure or a setting moved, and the parts that could not be compared.NoAdministrator, Auditor
aigov_propose_exposure_scanAsk for an exposure scan of an organization. It only proposes: nothing is read until an administrator approves it in the Portal. Asking again while a scan is open returns that scan.YesAdministrator

See AI exposure scan and Prospects.

People

ToolWhat it doesChanges somethingWho
aigov_list_staffThe staff list with the labels your name policy allows. No scores.NoAdministrator, Reviewer, Reveal approver
aigov_list_staff_groupsThe staff groups and how many people each holds.NoAdministrator, Reviewer, Reveal approver
aigov_set_staff_groupCreate a staff group, rename it, or set its members.YesAdministrator
aigov_list_identity_reviewsThe AI identities that are not yet matched to a person.NoReviewer, Administrator
aigov_dismiss_identity_reviewSay that an unmatched identity needs no match.YesAdministrator
aigov_request_revealAsk for a person's name to be shown, which a second person must approve in the Portal.YesReviewer
aigov_list_rolesThe governance roles people hold. No email addresses.NoAdministrator
aigov_list_data_requestsThe requests staff made about their own data.NoAdministrator

Sources

ToolWhat it doesChanges somethingWho
aigov_list_sourcesThe AI sources, what each captures and when it last reported.NoAdministrator, Auditor
aigov_get_source_healthWhat one source delivered and what was refused, hour by hour.NoAny role
aigov_list_vendor_connectionsThe vendor admin connections that vendor sources read through.NoAdministrator, Auditor
aigov_create_sourceCreate an OTLP source. The token is revealed in the Portal, never here.YesAdministrator
aigov_set_source_enabledSwitch one AI source off. Switching a source on is refused and sent to the Portal, because it can start collecting data.YesAdministrator

Rules

ToolWhat it doesChanges somethingWho
aigov_list_rule_packsThe rule packs and whether each is on.NoAny role
aigov_list_rulesThe detection rules, their mode, severity and actions.NoAny role
aigov_get_ruleOne rule in detail, with its versions.NoAny role
aigov_set_rule_packTurn a rule pack on, always in Test mode, or off. Turning off a pack that has a Live rule stops that rule's alerts and blocking, so it needs the typed confirmation DISABLE and the pack key.YesAdministrator
aigov_set_rule_overrideSet a rule to off or test, or change its severity. Never to Live. Taking a Live rule to off or test stops its alerts and blocking, so it needs the typed confirmation DISABLE and the rule key; the severity of a Live rule is changed in the Portal.YesAdministrator
aigov_create_custom_ruleCreate your own detection rule, always in Test mode.YesAdministrator
aigov_update_custom_ruleReplace the definition of one of your rules, in Test mode.YesAdministrator
aigov_replay_ruleReplay a rule over the last days and read how many records it would have matched.YesAdministrator
aigov_get_client_dictionaryThe client name dictionary and its last runs.NoAdministrator
aigov_set_client_dictionaryAdd an entry or an alias, or exclude an entry.YesAdministrator
aigov_get_jev_settingsHow AI judgments are set for your organization.NoAny role
aigov_list_reach_rulesThe rules for what AI agents may do in your connected tools, with how many calls each refused, held or would have refused or held in the last seven days. Never a person and never an argument.NoAdministrator
aigov_propose_reach_rulePropose an agent rule. It is always stored in Test mode and shows as proposed by an AI tool; an administrator makes it live in the Portal by typing ENFORCE RULE.YesAdministrator. Needs the Pro plan and a second factor.
aigov_get_reach_approvalWhether one held agent call is waiting, approved, rejected or expired, and until when an approved call can be repeated.NoAdministrator

Compliance

ToolWhat it doesChanges somethingWho
aigov_get_evidence_statusWhich evidence is recorded and what is missing.NoAny role
aigov_get_readinessHow ready your organization is for each framework.NoAny role. Readiness for one client needs Business.
aigov_get_policyThe current AI use policy, a version, or the draft.NoAny role. The draft needs Administrator.
aigov_list_policy_versionsThe versions of the AI use policy and how many people acknowledged each.NoAny role
aigov_draft_policyStart a draft of the AI use policy. It tells no one.YesAdministrator
aigov_save_policy_draftSave text into the draft. Publishing stays in the Portal.YesAdministrator
aigov_record_acknowledgementRecord that a person signed the policy on paper.YesAdministrator
aigov_list_trainingThe training requirements and how many people are covered, as counts.NoAny role
aigov_set_training_requirementRequire a course for all staff, a tool or a group.YesAdministrator
aigov_record_trainingRecord that a person finished an outside course.YesAdministrator
aigov_list_evidence_packsThe evidence packs asked for and their status. For an Administrator or an Auditor, each pack also says where its file is: with StackJack, or only in your own storage.NoAny role
aigov_request_evidence_packAsk for an evidence pack without names. You download it in the Portal.YesAdministrator, Auditor
aigov_list_auditThe changes made in AI Governance, newest first. When your records follow your own storage, it names the older days that are only there, and reads one of them back when you give its day, checked against the daily integrity record.NoAdministrator, Auditor
aigov_get_privacy_settingsJurisdiction, the staff notice, retention and whether collection is allowed.NoAdministrator, Auditor

Gateway and storage

ToolWhat it doesChanges somethingWho
aigov_get_gatewayThe in-path lanes, their block settings and the gateway budgets.NoAdministrator
aigov_list_gateway_keysThe personal gateway keys by id and scope. Never key text.NoAdministrator
aigov_revoke_keyStop a personal AI gateway key for good, by its id or its prefix. The AI gateway stops accepting it within about a minute.YesAdministrator
aigov_set_gateway_budgetCreate or change a soft monthly budget that warns and never stops a call.YesAdministrator
aigov_list_mcp_proxy_serversThe MCP servers behind the StackJack proxy. Never an address or a credential.NoAdministrator
aigov_get_browser_policyWhat the browser extension does on unapproved sites, personal accounts and uploads.NoReviewer, Administrator
aigov_get_endpoint_policyThe endpoint agent policy. Never the uninstall code.NoAdministrator, Auditor
aigov_list_endpoint_devicesThe endpoint agent fleet by version and health.NoAdministrator, Auditor
aigov_get_hooks_statusWhether the Claude hooks receiver works. Never its address or secret.NoAdministrator, Auditor
aigov_list_archive_targetsYour storages: the role (main store or extra copy), the one state and its sentence, and their health. Never a credential.NoAdministrator

The tools that change something are offered only when the access setting is Read and act. Where an AI tool changes a setting, it makes the safe change: a new rule or rule pack starts in Test mode, a source is switched off but never on again, and a draft tells no one. Anything that widens collection, sends data out or changes who sees data stays in the Portal.

Deploy and manage

These tools do what the Deploy and manage tabs do. The Pro plan is needed for all of them. The library checks your access in the client on every call, in the same words the Portal uses, so a tool never does more than your role may. The Administrator-only items stay for an Administrator of the whole organization. See Deploy and manage.

ToolWhat it doesChanges somethingWho
aigov_list_standardsYour standards: the current version, how many settings it sets, whether drift enforcement is on and how many clients run it.NoAny role that may see standards
aigov_get_standardOne standard: each setting with its value in words, what each version changed, and its clients with the values each client still has to give. A list of people shows as a number.NoAny role that may see standards
aigov_list_starter_standardsThe ready-made standards: each one's name, the Microsoft guide it follows, and how many settings StackJack sets and how many are done by hand.NoAny role that may see standards
aigov_list_pending_changesThe changes that wait for a decision that you may see, with what the client holds now and what it will be in words, the cost and who decides now.NoAny role. You see the steps your access covers.
aigov_get_driftWhere a client differs from its standards, why StackJack did not fix it, and the settings it could not read.NoAny role with access to the client
aigov_get_fleetEvery client you can see with where each standard stands there: how many settings match of how many StackJack read, how many differ, enforcement and the last check, and the same setting that differs at several clients. For an Administrator of your whole organization, it also lists the setup steps still missing, with where each is done.NoAny role. You see the clients your access covers.
aigov_list_seatsWho holds AI seats at a client, group by group. People are counted, not named.NoAny role with access to the client
aigov_get_partner_app_healthHow the Microsoft partner app and the service account are doing.NoAdministrator
aigov_get_deploy_runOne run: the summary of the plan, each step with what the client held when StackJack made the plan and what it will be, in words, its result, and each line to do by hand with its tick (including starting a sign-in job, with its link). A finished run also gives its result line. Set proof on a finished run to get the proof for the client. An automation that an approval rule names also gets facts for its decision: the standard, the client's seat usage and cost, and the client's earlier runs.NoAny role with access to the client
aigov_tick_checklist_lineTick, or take back the tick of, a setting that a person changes by hand in the vendor's console. The run keeps the time, and the proof for the client shows it. Only tick a line that the person told you is done.YesApprove changes up to a limit at the client, or Request changes on a run you planned yourself. An automation cannot use it.
aigov_preview_standard_runAsk StackJack to read a client and plan a standard. It returns a run id; the plan is ready in about a minute. Nothing is sent.YesRequest changes at the client
aigov_plan_standard_at_clientsAsk StackJack to read several clients and plan one standard at each, up to 25 clients at once (more is refused). A client that does not have the standard is skipped: assign it first with aigov_manage_standard. Nothing is sent; each plan still needs a submit and your approval rules.YesRequest changes at each client
aigov_put_back_settingAsk StackJack to plan putting one setting back to its standard at one client or several, up to 25 at once. Nothing is sent; each plan still needs a submit and your drift fix approval rule.YesRequest changes for drift fixes at each client
aigov_propose_standard_runSend a planned run for approval. Your approval rules decide each step.YesRequest changes at the client
aigov_decide_pending_changeApprove, reject or skip waiting changes. A person types the run's phrase in confirmation. An automation that an approval rule names decides only the steps handed to it, and types no phrase.YesA person with Approve changes up to a limit, or Apply changes on their own run, and a second factor. The one tool an automation may call to decide.
aigov_set_enforcementTurn drift enforcement on or off for one client or for a standard, or follow the standard.YesApply changes for enforcement at the client. A whole standard needs an Administrator of the whole organization.
aigov_propose_seat_changeAsk to take seats away from people at a client. It returns a run to read and send.YesRequest changes for seat changes
aigov_propose_purchaseAsk to buy seats through a distributor you linked. It returns a run with the quote and its currency.YesRequest changes for purchases
aigov_manage_standardCreate a standard, publish a version, archive, assign, remove an assignment, set or clear an override, set enforcement.YesAdministrator of the whole organization
aigov_copy_starter_standardCopy a ready-made standard into your library as a new standard. The copy starts with enforcement off: StackJack reports what differs and changes nothing until a person turns enforcement on.YesAdministrator of the whole organization
aigov_manage_approval_policyList, set or remove approval rules. A widening change needs ALLOW AUTOMATIC CHANGES and a second factor.YesAdministrator of the whole organization
aigov_manage_client_accessList, set or remove what a client role may do.YesAdministrator of the whole organization
aigov_manage_driftAccept a difference with a reason, end it, resume enforcement or ask for a check.YesApply changes for enforcement; Request changes for a check
aigov_manage_seatsWatch or stop watching a seat group, change its cleanup settings, keep or release one seat, resume a stopped cleanup.YesApply changes for seat changes
aigov_manage_purchase_settingsRead, set and reset the purchase safety stop, and set or remove the client's distributor account. Raising the limit needs ALLOW AUTOMATIC CHANGES and a second factor.YesAdministrator of the whole organization

An automation that an approval rule names reaches only three of these tools: aigov_list_pending_changes, aigov_get_deploy_run and aigov_decide_pending_change.

How answers look

A list comes in pages of up to 50 items, with a link to the Portal page that holds the rest. Text your own people typed, such as an incident summary or the reason for a request, is shortened in the answer, and the answer says so. Reading or deciding in an AI tool is not billed as tool usage. A decision is recorded in your audit log as made by the person who made it, exactly as it would be from the Portal.

Names

Staff names reach the AI tool exactly where your name policy shows them in the Portal. When names are pseudonymous, the AI tool receives the pseudonym, never the name, unless an approved reveal lets that person see it. No matched value or text of a prompt or an answer is ever sent to an AI tool. A finding shows its masked token, as the Portal does.

Asking to see a name from an AI tool only files the request. A second person approves or declines it in the Portal. The AI tool is never given the reason or the case reference that was typed with it.

What stays in the Portal

Some things stay in the Portal whatever the setting says: approving a name reveal, merging or splitting identities, pushing settings to endpoints, closing an incident, switching a source on, adding a person to a staff group that an alert policy or a rule names, putting a rule in Live mode, making or repairing the Microsoft partner app and signing in its service account, publishing a policy, downloading a report pack, exports and erasing a person.

Secrets never travel through a tool

No tool takes or returns an ingest token, a vendor key, a channel secret, an archive credential or a signing secret. Where a secret is needed, you create it in the Portal.

If your AI tool refuses

These are the answers a tool gives when it cannot help. Each comes with a fixed code the AI tool can read and, where a Portal page helps, a link.

  • AI Governance is not available: "AI Governance is not available for your organization."
  • Not set up yet: "AI Governance is not set up for your organization yet. Start it in the Portal."
  • Connected with an endpoint key or an automation: "AI Governance tools act for a person. Connect your AI tool by signing in to StackJack yourself; automations and shared endpoint keys cannot use them." and "Connect this AI tool by signing in to StackJack (not with an endpoint key) to use AI Governance."
  • The setting is Off: "Your organization has turned off AI Governance in AI tools. An administrator can change this in the Portal."
  • The setting is Read only and a decision was asked for: "Your organization allows AI Governance reads in AI tools, but not changes. Make this change in the Portal."
  • No AI Governance role: "You need a governance role for this. An administrator can grant one in the Portal." When you hold a role but not the one the tool needs, it says, for example, "You need the governance administrator role for this."
  • No recent second factor: "Sign in to StackJack again from this AI tool to use AI Governance: through your organization's single sign-on, or with a second factor (for example an authenticator app)." The AI tool only shows this sentence. It does not open a sign-in. Disconnect StackJack in the AI tool and connect it again, then sign in the way the sentence says.
  • A higher plan is needed: "This needs the Business plan or higher." The plan the feature needs is named. Some refusals say "This needs a higher plan."
  • Something moved since you read it: "That request has already been decided or withdrawn. Read it again before making a change." The matching sentences cover an incident or a finding.
  • A staff group that routes alerts: "That group is named by an alert policy or a rule scope, so adding a person to it would start sending that person's alerts, or applying a rule to them, with no one approving it. Add people to this group in the Portal." Removing people, and changing a group that nothing names, works from an AI tool.
  • A connector hostname that does not serve AI Governance: "This tool is not served on the hostname this connection is using." It is the answer every connector tool gives on a hostname made for another connector. Reconnect on your organization's primary StackJack endpoint.
  • Not found: "That record was not found in your organization." An alert, a request or an incident that is not in your organization, or not one your role may see, gets the same answer.
  • A fault on StackJack's side: "AI Governance could not complete this request. Try again in a moment." After a decision or a change, read the record again before asking twice.
  • A region that cannot take changes right now: "AI Governance cannot be changed right now in this region. Try again later."

Troubleshooting

What you seeWhat to check
The tools do not appear in your AI toolThe access setting is Off, you hold no AI Governance role, or you connected with an endpoint key.
Only read tools appearThe setting is Read only.
The AI tool shows "Sign in to StackJack again"Your sign-in is too old for AI Governance. The AI tool does not start a new sign-in. Disconnect StackJack in the AI tool, connect it again, and sign in with single sign-on or a second factor.
A tool says a plan is neededSee Plans and what each includes.

Next