Know every AI tool in use, and who approved it.
SAVI sees which AI tools people use from your workspace connectors, keeps a register of what you approved, and flags use that is not on it, by team.
Example workspace
| AI tool | Decision | Risk tier | Review |
|---|---|---|---|
| Microsoft 365 Copilot | Approved | Medium | Due Mar 2027 |
| GitHub Copilot | Approved with conditions | Medium | Due Jan 2027 |
| Claude | Approved | High | Overdue |
| Gemini | Under review | Unclassified | Due Dec 2026 |
| Perplexity | Not approved | High | Due Aug 2027 |
Illustrative data. Findings are by team; seeing the people behind one needs an admin permission, and each look is recorded.
A register built from use, not from a survey.
A register filled in once goes stale. SAVI keeps comparing yours with what people actually use.
01
Found from your systems
Activity reports and sign-ins from the tools you already run show which AI people use, with no code change.
Workspace connectors
02
Owners and review dates
Each tool records its purpose, data classification, risk tier, owner and manager, and a review that falls due each year.
Approved-AI register
03
Use that is not approved
Use of tools the register has not approved is listed by team. Names sit behind a permission, and every look is recorded.
Audited people view
04
Agents on the register too
The agents you build are registered and checked against their own calls for missing owners and drift.
Agent registry
See AI use from the systems you already have.
Connect the workspace tools your company already runs. SAVI reads their activity reports and sign-ins, names the AI tool each one is about, and says how sure it is.
Personal accounts on personal devices are invisible to every connector. SAVI never reads what anyone typed this way.
Workspace connectors
No code change
- Confirmed
Microsoft 365
Activity report
- Confirmed
Google Workspace
Activity report
- Seats
GitHub Copilot
Seat assignments only
- Confirmed
Claude Enterprise
Activity report
- Likely
Entra ID
Sign-ins to AI tools
- Likely
Okta
Sign-ins to AI tools
GitHub Copilot
Example entry
- Decision
- Approved with conditions
- Conditions
- Company accounts only; no client documents
- Purpose
- Code completion in the main repositories
- Data classification
- Internal
- Risk tier
- Medium
- Owner · manager
- eng.lead@example.com · cto@example.com
- Approval reference
- RISK-2026-114
- Last reviewed
- 14 Jan 2026, by risk.lead@example.com
- Review due
- 14 Jan 2027
An approved tool with no risk tier, no manager or an overdue review is flagged.
Decide once, and keep it current.
Record each tool's decision with its conditions, purpose, data classification and risk tier, who owns it and who signed it off. SAVI tracks when each review falls due and flags the gaps.
The register is your decision, kept with its history. SAVI does not approve tools for you.
Unapproved use, by team.
Each finding is one tool in one team, with its severity taken from the register: high for a tool you have not approved, medium for one you have not decided on. Acknowledge, resolve or dismiss it.
A finding is not a verdict. Seeing the people behind it needs an admin permission, and every look is recorded in the audit trail.
Keep your own agents on the register.
Register the agents you build with their owner, environment and data classification. SAVI compares the register with their calls and lists what does not match.
Governance rules can then treat agents that are not registered differently.
Agent registry · last 30 days
Example data
2
Unowned
Registered, but nobody owns them
3
Not registered
Seen in calls, missing from the registry
4
Dormant
Registered, no calls in 30 days
1
Drift
Declared public, personal data seen
Summarise a customer complaint
Approved · v3Summarise the complaint below in three sentences. Name the product, the problem and what the customer asked for. Do not include names or contact details.
Personal data in a draft is flagged as it is typed.
Share prompts that are approved.
Publish prompts people can copy into their AI tools, with versions you can go back to. SAVI records who copies each one, which also feeds the adoption score on the spend and value page.
How the register works.
Sources, limits, privacy and reviews.
Where does SAVI see AI use?
From connectors to Microsoft 365, Google Workspace, Entra ID, Okta, GitHub Copilot and Claude Enterprise. An activity report from the vendor counts as confirmed use; a sign-in to the AI tool counts as likely. GitHub Copilot reports seat assignments only. No code change is needed.
What can it not see?
Personal accounts on personal devices are invisible to every connector, so they never appear. The connectors report activity, not what anyone typed, so prompts are never collected this way.
Who can see which people are behind a finding?
Findings are shown by team. Seeing the people behind one needs an admin-only permission, and every time someone looks, it is written to the audit trail, because it names employees.
Does a finding mean someone broke a rule?
No. A finding says a tool was used that the register marks as not approved, under review, or not yet decided. It is a prompt to review, which you can acknowledge, resolve or dismiss.
How do reviews work?
Each entry records who reviewed it and when. A review falls due a year after the last one, or after the entry was created if it has never been reviewed. Approved tools with an overdue review, no risk tier or no manager are flagged. Every change is in the signed audit trail.
What is checked for agents?
Registered agents are compared with what their calls show: agents that are unowned, agents seen in calls but not registered, agents with no recent calls, a different environment from the one declared, and an agent declared public whose calls contained personal data. Nothing is changed for you; each is listed for review.
How does the prompt library work?
Prompts are drafts until published. Editing saves a new version, and an older version can be made current again. Personal data is flagged as a draft is typed. Each copy is recorded with the person, and the audit trail records a fingerprint of the text, not the text itself.
Turning the register into rules? Governance · Agent tool approvals
See what AI your teams are really using.
Connect a workspace tool in the demo and we will walk through the register, the findings and the reviews.