TekuBase AIGuard

National AI use control · production edition

AI use control, upload protection, and audit evidence for institutions.

Register institutions, generate secure deployment kits, inspect AI use, and export audit evidence without storing raw sensitive content.

Loading system status…

Command Ledger

Ready to refresh.

Loading…

Create institution

Institution ID Name Sector Risk tier

No institution action yet.

Invite the first institution administrator

Create the institution workspace first, then send password-setup instructions to its verified administrator. Public self-signup remains disabled.

Administrator email Administrator name

No administrator invitation sent.

AI Tool Registry

The approved inventory for websites, desktop applications, local model runners, APIs, and browser extensions.

Institution ID Tool ID Display name Tool type DecisionNew approvals start in Discovery Queue and require a second reviewer. Domains Comma-separated, no paths Executable names Publishers Package IDs Data handling

Tool	Type	Decision	Coverage
Refresh the registry to load tools.

Technical result

No registry action yet.

Discovery Queue

New websites and applications reported automatically by managed browsers, endpoints, gateways, DNS, proxies, and firewalls.

Default action: an unknown AI destination is blocked and held for review. Only destination and application metadata is stored.

Discovered item	Source	Status	Action
No discoveries loaded.

Review selected discovery

Discovery ID Decision Tool ID Tool name

Controlled discovery simulator

Use synthetic metadata only. Managed devices normally report this automatically.

Institution IDTypeDetected domainDetected application

Technical result

No discovery action yet.

Managed Devices

Enroll institution computers and monitor Browser Guard, Endpoint Agent, application discovery, and policy status.

Institution ID Device name Platform

One-time enrollment secret

Copy it directly into the approved device configuration. It cannot be displayed again and is never stored in raw form.

Device	Site / group	Platform	Access	Connection	Action
No managed devices loaded.

Selected device control

Device IDAccess decision

Replacement device credential

Install it directly on the selected device. It is shown once and is never stored in raw form.

Technical result

No device action yet.

Fleet Management

Organize institution sites and device groups, then issue short-lived enrollment campaigns for controlled bulk installation.

Unique device identity: a campaign authorization is temporary and usage-limited. Every computer exchanges it once for its own revocable credential.

1. Institution site

Institution ID Site ID Site name Site type

2. Device group

Group ID Group name Platform

3. Enrollment campaign

Maximum devices Valid for hours Installer platform

One-time campaign authorization

Use only in the approved deployment system. Revoke it after rollout. The raw value cannot be displayed again.

Campaign	Scope	Usage	Expires	Status	Action
No enrollment campaigns loaded.

4. Central device actions

Safe remote administration: choose a site, group, selected devices, or every matching device. Endpoints accept predefined actions only—arbitrary PowerShell and raw credentials are never sent.

Scope Action Batch size Maintenance window

Select	Device	Site / group	Connection
Refresh fleet to load devices.

Action	Scope	Progress	Status	Approval
No central actions loaded.

Technical result

No fleet action yet.

Institution Deployment Center

Generate institution-specific gateway, endpoint, browser guard, policy bundle, installation docs, checksums, and a downloadable ZIP package.

Institution ID Institution name

No deployment kit generated in this session.

Operator access

Allowlist an existing Authentication user and assign one controlled role. Institution-scoped roles require at least one institution ID.

Email Display name Role Status Institution IDs

No operator action yet.

Approvals

Institution ID Request type Subject reference Justification

Request	Subject	Status	Action
Refresh approvals to load requests.

Record a decision

Approval ID Decision Decision summary

No approval action yet.

Incidents

Institution ID Incident type Severity Redacted summary Detail for hashing

Incident ID Status Status summary

No incident action yet.

Prompt Guard

Institution ID User ID

Prompt text

No prompt inspected yet.

Document Upload Guard

Managed browsers detect the destination automatically. Operators do not maintain URLs during normal use.

Automatic coverage: every available external file selection is paused and inspected before the destination page receives it. Some external services disable attachments for signed-out or restricted accounts; use this controlled simulator when their file picker is unavailable.

Institution ID User ID Destination capture Test destination profileUsed only by this controlled simulator. File name

Document/content sample

No upload inspected yet.

Developer Guard

Institution ID Repository

Code sample

No code scanned yet.

Evidence Export

Evidence scope Institution ID

No evidence action yet.

Help Center

Choose the page you are working on. Each guide explains who may use it, exactly what to click, what success looks like, and what to check when it fails.

New institution? Follow this order:

Institution Setup
Operator Access
AI Tool Registry and Discovery Queue
Fleet Management
Policy Control
Deployment Center
Managed Devices
Evidence

Command Ledger System status and tamper-evident activity

Use: Operators and auditors. Start here each day.

Select Command Ledger.
Press Refresh.
Confirm the counts, then confirm Audit chain: PASS.
Review the newest events in the technical result.

Success: A current refresh time, expected counts, and PASS. If unchanged: wait for the action to complete, refresh once, and check sign-in/network status. A refresh does not create an event.

Institution Setup Create a workspace before inviting users

Use: National operator only.

Enter a stable lowercase Institution ID; do not use a person’s name.
Enter the official institution name and approved region.
Create the workspace.
Invite its verified administrator using an official email address.

Success: the institution appears in scope and the administrator receives password-setup instructions. Public self-signup is intentionally disabled. Never share temporary credentials by chat.

AI Tool Registry Define approved, monitored, and blocked tools

Use: Institution administrator or security officer.

Search the registry before adding a duplicate.
Enter the canonical tool ID/name and reliable domain, executable, publisher, or package indicators.
Choose Approve, Monitor, Block, or Review required.
Save; complete independent approval when requested.

Success: one unambiguous registry row with the intended decision. Saving does not activate endpoints until a policy release is published.

Discovery Queue Review unknown AI destinations and applications

Use: Security officer.

Refresh discoveries and select Review on one pending item.
Verify the domain, executable, publisher, package, and source.
Choose Approve, Monitor, or Block and add a defensible reason.
Complete second-person approval if required, then publish policy.

Success: status becomes complete and the registry contains the tool. Unknown AI remains blocked and held for review; metadata only is stored.

Managed Devices Check identity, access, agent version, and connection

Use: Institution IT and security.

Refresh devices.
Confirm the device is ACTIVE and ONLINE with the expected site/group and agent version.
Select a device only for individual disable/enable or controlled recovery.
Use Fleet Management for health checks, rotation, updates, rollback, and bulk work.

Statuses: NEVER_CONNECTED means no heartbeat; STALE means the last heartbeat is old; DISABLED means cloud access is revoked. Do not rotate manually when automatic fleet rotation is available.

Fleet Management Sites, groups, campaigns, and central actions

Use: Institution IT; security approval for high-impact work.

Create a site, then a device group.
Create a short-lived campaign with capacity matching the rollout batch.
Deploy the verified enterprise package with Intune, Group Policy, or approved endpoint management; every computer exchanges the campaign authorization once for its own identity.
Revoke the campaign after enrollment.
For operations, select a scope/action, use a maintenance window, create the controlled action, and refresh progress.

Success: enrolled devices become ONLINE and jobs reach COMPLETED. Multi-device update, rollback, disable, uninstall, and rotation require an independent approver. Endpoints never accept arbitrary remote PowerShell.

Policy Control Preview and publish the effective enforcement policy

Use: Institution administrator/security officer.

Enter the institution ID and preview the effective policy.
Confirm approved, blocked, and monitored counts plus raw sensitive storage Disabled.
Resolve pending approvals or discoveries.
Publish a numbered policy release.
Verify managed devices receive the new release.

Success: a new active version with an audit hash. Preview is read-only; publish is the step that makes registry decisions deployable.

Operator Access Invite people with least-privilege roles

Use: National or authorized institution administrator.

Choose the narrowest role and institution scope.
Use the person’s verified official email.
Send the invitation/password-setup email.
Ask the user to sign in and verify only their assigned scope.

Recovery: use Forgot password; responses intentionally do not disclose whether an account exists. Never create shared operator accounts.

Approvals Independent review for high-impact decisions

Use: A different authorized security officer—not the requester.

Refresh and select the pending request.
Verify subject, action, device count, scope, justification, and maintenance window.
Choose Approve or Reject and enter a specific summary.
Record the decision, then verify the linked job or tool status.

Success: the request and linked operation show the same final state. Never approve only because the requester is waiting.

Incidents Record, assign, and close security cases

Use: Security operations.

Create an incident from verified metadata; do not paste raw prompts or documents.
Set severity, owner, and containment state.
Link relevant audit/event identifiers.
Record remediation evidence before closure.

Success: accountable owner, timestamps, linked evidence, and a justified final status. Disable affected devices when compromise is suspected.

Deployment Center Generate verified institution installation kits

Use: Authorized deployment operator.

Select the institution and generate its deployment kit.
Download the ZIP and record its SHA-256 separately.
Verify integrity before installation.
For production, sign scripts/MSI with the organization’s trusted code-signing certificate.
Distribute through Intune, Group Policy, or approved endpoint management.

Safety: packages must not contain campaign tokens, device credentials, signing keys, PINs, or passwords. Pilot unsigned packages require checksum verification; production packages should be signed and timestamped.

Prompt Guard Inspect text before it reaches an AI service

Use: Security testing and controlled troubleshooting.

Use synthetic text only.
Select the correct institution/user/destination context.
Run inspection and read decision, classification, and reason.
Confirm an audit event exists without raw prompt storage.

Expected: public content follows tool policy; credentials, identity references, and restricted data are blocked. Do not use the simulator to process real confidential data.

Upload Guard Inspect file metadata/content before external upload

Use: Security testing and managed Browser Guard.

Use a controlled synthetic test file.
Choose the destination and classification context.
Inspect before upload.
Verify restricted content is blocked and the audit record contains no raw file.

Troubleshooting: a website’s own Add files button may require login; that is separate from AIGuard. Browser Guard pauses HTTPS uploads and applies policy when an upload is attempted.

Developer Guard Prevent unsafe code and secrets from leaving the institution

Use: Developers and security reviewers.

Use synthetic code in testing.
Inspect the intended AI destination and code snippet.
Review detected secrets/unsafe patterns and the allow/block decision.
Remove or rotate exposed credentials before retrying.

Safety: never paste production keys, tokens, citizen data, or proprietary source merely to test detection.

Evidence Export verifiable audit material

Use: Auditor or authorized operator.

Choose the institution and approved date/scope.
Verify Audit chain PASS before export.
Generate and download the evidence ZIP.
Record the checksum and store the package in the approved evidence repository.

Success: downloadable metadata/evidence with verifiable hashes and no raw prompts, files, code, passwords, or credentials. Do not delete historical events to clean the dashboard.

Frequently asked questions

Short answers for the questions operators and institution IT teams are most likely to ask.

Why can’t an institution sign itself up?

Public self-signup is disabled to prevent unverified organizations from creating trusted workspaces. A national operator creates the institution and invites a verified administrator.

Does “Institutions 3” mean one institution owns three others?

No. It is the number of institution workspaces visible to the signed-in operator. National operators can see several; institution-scoped users see only their assigned IDs.

How do we install AIGuard on 500 computers?

Create sites and groups, build and verify the enterprise package, then deploy it in staged batches through Intune, Group Policy, or equivalent endpoint management. A limited campaign authorizes enrollment, but every computer receives its own unique revocable credential.

Do IT staff paste a credential on every computer?

No. The one-time campaign authorization is supplied securely by the deployment system. Each endpoint exchanges it once, protects its own device credential with Windows DPAPI, and never returns the raw value to the dashboard.

What do ONLINE, STALE, NEVER_CONNECTED, and DISABLED mean?

ONLINE recently sent a heartbeat. STALE connected before but is overdue. NEVER_CONNECTED enrolled but has not sent its first heartbeat. DISABLED has had cloud access revoked by an operator.

Why did Refresh appear to do nothing?

Refresh reads current cloud state; it does not create a new event. Watch the “Last refreshed” message, confirm you are signed in, then wait for the device or approval operation and refresh once more.

Why is a tool still blocked after I approved it?

The discovery decision may still need an independent approval, and the effective policy must be published. Then the endpoint must refresh its policy. Check Approvals, Policy Control, and the device heartbeat in that order.

Why can’t I approve my own request?

High-impact actions use two-person control. The requester must not be the reviewer. A different authorized security officer verifies the scope and records Approve or Reject.

Can central IT run any PowerShell command remotely?

No. Endpoints accept only predefined, schema-validated AIGuard actions such as health check, policy refresh, rotation, update, rollback, disable, and uninstall. Arbitrary remote commands and raw credentials are intentionally rejected.

Should we use Rotate device credential manually?

Normally no. Use Fleet Management’s automatic rotation. Manual rotation is a controlled recovery path only when the endpoint cannot authenticate and requires securely applying a replacement locally.

Why does an upload or prompt get blocked on a monitored tool?

The tool decision and content decision are separate. Monitoring can allow ordinary public content while DLP still blocks credentials, identity references, restricted documents, secrets, or unsafe code.

Does AIGuard store our prompts and files?

No raw sensitive prompt, document, code, password, token, or credential content is stored. The ledger keeps controlled metadata such as decision, classification, actor, timestamps, hashes, and audit-chain links.

Can we use an unsigned package?

A checksum-verified unsigned package is acceptable only for a controlled pilot under your security process. Production distribution should use an organization-validated code-signing certificate, trusted timestamping, and signature verification.

What should we do if a laptop is lost?

Select the device in Managed Devices and disable it centrally. Investigate and preserve evidence. If recovered and trusted, follow the authorized recovery process; do not simply create a duplicate identity.

Where should I start when something fails?

Check Command Ledger, the page’s Technical result, Managed Devices connection status, pending Approvals, active Policy version, and Incidents. Record exact IDs and error messages—never send secrets or raw institutional data to support.

Governance and rollout reference

Follow this guide in order for a controlled institution rollout. Public self-signup is disabled: every institution, operator, device, policy, and high-impact action remains inside an approved administrative workflow.

Start Roles Fleet Operations Signing Recovery

1. Onboard an institution

A national operator creates the institution workspace first, then invites its verified administrator. The administrator receives password-setup instructions directly by email; AIGuard never stores a password or reset link.

Open Institution Setup and use a stable institution ID.
Invite the institution administrator at an official, verified email address.
Use Forgot password for recovery; the response does not reveal whether an account exists.

2. Assign the correct role

The number shown at the top is the count of institution workspaces in scope. It does not mean one institution owns another. A national operator may administer several institutions; institution roles see only their assigned institution IDs.

National operator: cross-institution control and onboarding.
Institution administrator: sites, groups, devices, and operators in its institution.
Security officer: reviews tools, incidents, approvals, policies, and fleet actions.
Auditor/developer reviewer: read or specialist review access only.

3. Register AI policy

Unknown destinations are blocked and held for review. Managed browsers, endpoints, gateways, DNS, proxies, and firewalls report destination and application metadata automatically.

Review discoveries as Approve, Monitor, or Block.
New approvals require an independent second reviewer.
Publish a numbered policy release after approved changes.

4. Organize the computer fleet

Create sites and device groups before enrollment—for example, head office, branch, data center, staff computers, security workstations, or developer devices.

1 computer: controlled PowerShell installation is acceptable for a pilot.
Up to 50: use a limited campaign with Group Policy or approved IT deployment.
Hundreds: use Intune or equivalent endpoint management with staged groups.
Every computer exchanges campaign authorization once for its own revocable identity.

5. Deploy safely at scale

Create a short-lived, capacity-limited enrollment campaign for one site and group. Keep its authorization outside MSI properties, source control, email, browser storage, and shared folders.

Begin with a small pilot and verify ONLINE heartbeats.
Roll out in batches using the provided Intune, Group Policy, or PowerShell profiles.
Revoke the campaign after the expected enrollment count is reached.
Browser Guard is distributed through managed Chrome/Edge enterprise policy.

6. Operate devices centrally

IT can target checked devices, a group, a site, or all matching institution devices. Selection scope applies to the filtered target, not only a convenient manual copy-and-paste workflow.

Available actions are health check, policy refresh, restart, automatic credential rotation, staged update, rollback, disable, and controlled uninstall.
Endpoints accept predefined, schema-validated actions only—never arbitrary remote PowerShell.
Offline devices receive unexpired queued work when they reconnect.
Use maintenance windows, small batches, progress refresh, pause/retry procedures, and rollback evidence.

7. Approve high-impact actions

Multi-device credential rotation, update, rollback, disable, and uninstall require two-person control. The requester cannot approve or reject their own request.

A second security officer reviews the action, scope, count, and maintenance window.
Rejected jobs remain at zero progress and never reach endpoints.
Approval and job outcomes are appended to the audit chain.

8. Rotate credentials automatically

The endpoint generates the replacement locally, sends only its SHA-256 verifier to the control plane, and protects the raw value with Windows DPAPI LocalMachine.

Administrators never view, copy, email, or paste the replacement.
Verify the job reaches COMPLETED and the device remains ONLINE.
Manual rotation is reserved for controlled recovery only.

9. Inspect AI use and evidence

Prompt Guard, Upload Guard, and Developer Guard classify content before external AI use. The ledger retains decisions, classifications, hashes, actors, and chain links—not raw prompts, files, code, passwords, or credentials.

Verify the audit chain after material operations.
Export evidence for internal audit, INSA readiness, management, and regulator review.
Use synthetic data for acceptance tests.

10. Sign production releases

Controlled pilots may use checksum-verified unsigned packages. Production distribution should use an organization-validated code-signing certificate issued to BookAddis Technology PLC.

Prefer a trusted hardware token or managed cloud-signing service.
Keep the private key non-exportable; never send the key, PIN, or password through chat, email, ZIP, source control, or AIGuard.
Run signing-readiness, sign PowerShell/MSI artifacts, timestamp signatures, and verify signatures before publishing.

11. Update, recover, and revoke

Use only integrity-verified enterprise bundles. The transactional updater backs up the current runtime, preserves device identity, checks health, and restores the backup if validation fails.

Use rollback only through an approved maintenance action.
Disable a lost or retired device centrally before local uninstall.
Rotate a suspected credential automatically; use manual recovery only when the endpoint cannot authenticate.
Never delete audit history to make a dashboard look clean.

12. Daily operator routine

Keep operations short and repeatable.

Review offline/stale devices and pending discoveries.
Review pending approvals and incidents.
Confirm policy release and fleet-job progress.
Verify the audit chain and export scheduled evidence.
Investigate repeated failures before retrying an action.

Operational rollout checklist

Step	Owner	Output
Approve operating scope	Executive sponsor + security office	Institution, users, risk scope, and success criteria
Create workspace	National or institution operator	Institution record and policy context
Invite administrator	National operator	Verified institution account with scoped role
Define sites and groups	Institution IT	Approved fleet topology and rollout scope
Generate kit	AIGuard operator	Integrity-verified ZIP, SHA-256, and installation guide
Create campaign	Institution IT + security	Short-lived, capacity-limited enrollment authorization
Enroll devices	IT endpoint team	Unique device identities and online heartbeats
Enforce network policy	Network/security team	Unapproved AI bypass paths blocked or monitored
Run acceptance tests	Security + department users	Allowed public prompt, blocked restricted upload, blocked risky code
Test central operations	IT + second approver	Health, rotation, rejection, reboot, and rollback evidence
Export evidence	Auditor/operator	Evidence ZIP and verified audit chain

Security rule: Do not email enrollment authorizations, credentials, signing keys, PINs, or certificate passwords. Do not upload real citizen, customer, credential, or confidential data during first validation. Use controlled synthetic data until the institution security office approves production traffic.