Frequently asked questions

Enterprise AI tools only govern prompts sent through their own interface. They do not cover what happens when an employee uses ChatGPT from a personal account, tries Claude, or picks up a brand-new AI tool. Proxara covers that gap, and works alongside the existing enterprise AI stack.

The device agent ships through the MDM the firm already runs (Intune, Jamf, Kandji, Workspace ONE) and activates itself on first boot; employees do nothing. Proxara provisions and runs the backend environment, so there is no infrastructure for the firm to build.

They keep using their preferred AI tools as before. Sensitive data is silently replaced before the prompt reaches the AI tool. The AI still gives a useful answer. The employee gets the real values back. Most of the time, they will not notice it is there.

ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Grok, and the rest of a registry of a thousand-plus AI sites and apps that refreshes continuously, with no list for the firm to maintain. The device agent captures AI traffic through the operating system, so browsers, desktop AI apps that honour the system proxy, coding tools, and API calls pass one inspection point, and anything the agent cannot inspect is reported rather than hidden.

Microsoft sign-in and identity traffic is never touched, and coverage of Microsoft AI surfaces follows the same registry as everything else, alongside the controls Microsoft provides inside its own tenant. Most of the exposure firms worry about sits outside that tenant: consumer AI accounts, native desktop apps, coding tools, agents, and MCP servers. Proxara governs that lane.

Analysis runs on a local model inside the environment provisioned for the firm, never a third-party AI service, and Proxara maintains that infrastructure without taking the data out of it. What reaches an AI provider is the redacted prompt. The mapping between real values and tokens stays in that environment, in the identity vault, where a reveal is possible only on the record.

It is a local model running inside the environment provisioned for the firm. It retains no inputs or outputs, learns nothing from the firm’s data, and nothing about the analysis leaves that environment.

Work is never blocked. If classification cannot finish in time, the prompt goes through unredacted by default and the degraded coverage is reported rather than hidden, so compliance can see when and where protection thinned. A firm that prefers to hold traffic in that state can set fail-closed instead, and automated agent traffic is always held.

No. Redaction is silent and applies to everyone evenly, and escalation is reserved for what firm policy marks as material. The intelligence is about tools, spend, and coverage rather than rankings of people. Most days, nobody opens a console at all.

Proxara does not alter the books and records of the firm. Sensitive data is redacted before prompts reach external AI, and the interaction record can route to the existing archive, which remains the system of record.

The archive stays the system of record, and Proxara sits alongside it. An archive connector can deliver the interaction record to the archive the firm already runs, over the ingestion API of that provider and under the archive credentials the firm already holds. It stays off until the firm configures it.

Yes. Configure sensitivity thresholds, add firm-specific vocabulary rules, and build custom policy packs on top of pre-built SEC and FINRA rule sets. A firm can also load the AI policy it already has and see those rules enforced live.