What you can control
An organization controls Verbiqo two ways, both pushed with the device-management tools it already has: managed settings, which set and lock values in Settings, and a signed privacy policy, which the engine enforces and which cannot be edited on the computer.
Settings and policy: which does what
| Managed settings | Signed privacy policy | |
|---|---|---|
| What it does | Fills in Settings and locks what you set: greyed out, Managed by your administrator. What you leave out stays each person’s own. | Checked by the engine on every request. Anything it does not allow is refused: another endpoint or model, a feature, a system, data it blocks. |
| Controls | The AI service, model, timeout, temperature, context window, Trusted CA, API key on or off, the embedding service for Prometheus, the licence file, extension limits, where the access record is sent, and switches | Which AI service and models, features and systems are allowed, and what data may be sent |
| Pushed as | A configuration profile (Mac) or registry values (Windows) | A signed file, pointed to by managed settings |
| If someone works around it | Nothing else stops them: a value only locks the field in Settings | The request is refused, from the app or the command line |
Use both for the AI service: managed settings so every computer is set up correctly and nobody has to type an endpoint, and the policy so nothing else is ever used. Use the policy alone for features, systems and data rules. Both are removable by someone with administrator rights on the computer, as with any setting your device management pushes; ordinary users cannot change either.
What you can control
| To | Use | Where |
|---|---|---|
| Set the AI service everyone uses: provider, endpoint, model | Managed settings | Deploy and manage |
| Refuse any other endpoint, protocol or model | Privacy policy: inference | Privacy policy |
| Set timeout, temperature and context window | Managed settings | Deploy and manage |
| Allow only some of PostgreSQL, MySQL, SQL Server, Redis, Elasticsearch, Kubernetes, Prometheus, Splunk and the security review, or none | Privacy policy: extensions | Privacy policy |
Allow only some of Rewrite, Explain, Ask, Summarize Document, and the security hunts in logs (security-logs) | Privacy policy: features | Privacy policy |
| Limit how hard Verbiqo works your databases and monitoring: queries and time per question, questions per hour | Privacy policy: limits, or managed settings | Protect your systems |
| Give different teams or people different systems, features and limits | A policy per group, targeted by your device management | Different rules for different groups |
| Block or redact secrets, personal and financial data, network identifiers, classification markings | Privacy policy: rules | Privacy policy |
| Turn off: reading other systems, the text around a selection, the clipboard fallback, update checks | Managed settings | Deploy and manage |
| Keep a record of what was allowed, blocked or redacted, with no content | Privacy receipts | Privacy policy |
| See every statement, command and request Verbiqo sent to your systems, send the record to your SIEM, and require it before anything is read | The access record, always on; managed settings auditForward and auditRequired | Access record |
| Licence a team without handing anyone the file, or computers that never reach the internet | Managed settings: licensePath, with a licence key or a seat file | Licences for an organization |
| Set the embedding service for Prometheus questions (a LiteLLM gateway, or vLLM behind it) | Managed settings: embeddingEndpoint, embeddingModel, embeddingUsesModelKey | Roll out to a team |
A locked value shows as Managed by your administrator in Settings and cannot be changed there. A policy that is deployed but cannot be read or verified stops requests to the model and every connected system, rather than letting them through.
Not available yet
| Not yet | Instead, today |
|---|---|
| Signing in with your identity provider (Keycloak, Entra ID, Okta) to the AI service | A static API key per computer or per person, kept in the Keychain or Credential Manager. Put identity and per-person keys at your AI gateway (LiteLLM supports both). |
| Windows integrated sign-in (Kerberos) to databases | A database account with a password. PostgreSQL and Splunk directory (LDAP) accounts work, since the server checks the password. |
| A central console or per-person usage reports from Verbiqo | Policies and settings are files and values your device management pushes. Usage and identity are recorded by your AI gateway; privacy receipts record decisions per computer. |
| Turning off individual rewrite modes | Rewrite is allowed or not as a whole. |
| A signed Windows installer | The Windows installer is not yet code-signed, so SmartScreen warns, and AppLocker or WDAC may need an exception. |
Who did what
Verbiqo keeps no record of what anyone asked: no history of questions or answers, on the computer or anywhere else. Where an organization needs to know who used the AI service and how much, the place to record it is the AI gateway every request passes through, with a key or sign-in per person. Verbiqo’s privacy receipts add, per computer, which requests the policy allowed, blocked or redacted, without their content. The access record adds every read Verbiqo made of a connected system: the statement, command or request as it was sent, by which account, on which computer, and whether the model or Verbiqo asked for it; never the question or the answer.
Next
Deploy and manage settings
Install on every computer and set the AI service, the embedding service and the licence. Roll out to a team
Write and sign a policy
Decide what Verbiqo may use and send. Privacy policy
Collect the access record
Decide where each computer sends its record of reads, and whether reads must be recorded first. Access record
Test it
Check each control on one Mac and one Windows PC before rolling out. Test a deployment