AI governance for managed service providers

See Every AI Tool.
Stop Every Leak.
Prove It.

Proctored is the agent you deploy to every client device to see which AI tools people use, stop sensitive data before it leaves, and prove it to auditors. Every client, one console.

Request a DemoSee How It Works

No credit card. No network changes. A thirty-minute demo on a real device, not slides.

Installs from your RMM in minutesmacOS, Windows, LinuxNothing on the client's network
Alerts · Tooth & Nail Dentallive
Payment card number blocked to ChatGPTnurse1 on CLINIC-WS-01 · 2 min ago
CRITICAL
Personal account used with Claude[email protected] on CLINIC-WS-07 · work accounts only
MEDIUM
New MCP server: GitHub (local)DEV-MBP-02 · Claude Code · reach: high
MEDIUM
Exception requested: customer list → Notionm.ortiz · 4 hours · "reconciling Q3 renewals"
DECIDE
Coverage regressed: Geminiinspected last week, now uninspected
HIGH
Prctored
Blocked by Tooth & Nail Dental's AI policy
Your message to ChatGPT was blocked
Payment card number

It contained data your organization doesn't allow to be sent to AI tools.

Request an exceptionGo back
Fits your stackConnectWise PSAAutotaskHaloPSANinjaOneDatto RMMN-ableMicrosoft TeamsSlackEntra IDGoogle Workspace
The problem you own

Your Clients' Staff Are Already Using AI. On Whose Terms?

You manage the endpoint, so when a customer list ends up in a chatbot, the question lands on your desk. Most tools can't even see it happen.

01

It's Not Just ChatGPT in a Browser

Desktop apps, coding assistants at the command line, MCP tool servers plugged into Claude Code and Cursor, and local models. Browser-extension tools see one of those.

02

Personal Accounts Are the Leak

A work laptop signed in to a personal ChatGPT account has no data agreement, no retention control and no audit trail. Nobody knows until something surfaces.

03

"We Have a Policy" Isn't Evidence

HIPAA, PCI DSS, CMMC and SOC 2 questionnaires now ask about AI. Your clients' auditors want evidence that the control ran, not a PDF that says it should.

How it works

Checked on the Device, Before Anything Leaves It

The agent routes only AI services through itself. Everything else on the device goes direct and is never touched.

Deploy From Your RMM

One installer per platform for all your clients. Push it with the client ID; devices enrol and report within a minute. Updates install silently and are signature-verified.

Proctored-Setup.exe install --org-key GOV-… --client-id TOOTHNAIL-1

Every Prompt Is Checked Locally

Prompts, uploads, pasted files and the text inside images are inspected on the device with the client's rules. The decision is made before the data leaves.

ALLOWWARNREDACTBLOCK

Run Every Client From One Console

Policies, approved apps, exception requests, alerts and client reports in one place. Alerts open tickets in your PSA and close themselves when resolved.

See

Every AI Tool, Down to the Product. Every Account.

Proctored tells Claude from Claude Code and ChatGPT from Codex, even on shared domains. It inventories desktop apps, local models and the MCP servers configured in coding tools. And it knows whether each person is on a work account or a personal one.

  • 162 AI apps, embedded AI features and web channels, each Approved, Unapproved, Blocked or Not managed per client
  • Coverage evidence per app from real devices, not a checkbox
  • "Work accounts only" refuses personal accounts at the prompt
proctoredby.com/console · Applications
The Proctored Applications page: each AI app with category, risk, requests in 30 days, blocked count, coverage evidence, accounts seen and the client's status
The Applications page, as it is: coverage evidence from real devices, accounts seen, and the client's decision per app.
Control

Sensitive Data Stops at the Device

Forty-three detectors with checksum validation and context, seven document types, custom lists and exact data match. Rules decide per client and per app: monitor, warn with a business reason, redact in place, or block. Employees who hit a block can ask for a time-boxed exception; you approve it in one click, and the grant works on that one device only.

  • Secrets, PHI, card numbers, CUI, customer records, contracts, HR and medical records
  • Word, Excel, PowerPoint, PDF and the text inside screenshots, read on the device
  • Prompt injection and jailbreaks caught at the same inspection point
Policies · Tooth & Nail Dentalbaseline + client rules
Block credentials and secretsBLOCK
API keys, private keys, passwords, connection strings · every app · no exceptions
PHI only to approved apps, after a warningWARN
MRN, NPI, ICD-10, medical records · unapproved apps blocked
Redact government IDsREDACT
SSN, passport, driver's licence · sent with the data removed
Customer list (exact match, 4,210 rows)BLOCK
hashed on upload · exceptions allowed · 2 pending
Prove

Evidence an Auditor Can Verify, Not a Slide That Says "We Have a Policy"

Every client gets a report with usage, protection, shadow AI, coverage and control evidence mapped to eight frameworks. Snapshots are frozen with a SHA-256 digest; the audit log is hash-chained and verifiable in the console; a read-only API feeds your SIEM.

  • HIPAA · PCI DSS · CMMC · SOC 2 · ISO/IEC 42001 · NIST AI RMF · GDPR · GLBA
  • A coverage alert when an app that was inspected goes by uninspected
  • Monthly snapshots generated on the first, ready for the quarterly review
AI Governance Report · SeptemberHIPAA control evidence
164.308(a)(1) Risk managementShadow AI inventory · 2 unapproved apps found, 1 blocked
EVIDENCED
164.312(a)(1) Access controlWork accounts only on ChatGPT · 3 personal accounts refused
EVIDENCED
164.312(b) Audit controls1,284 events · audit chain verified
EVIDENCED
164.312(e)(1) Transmission securityPHI blocked to unapproved AI apps · 7 events
EVIDENCED
snapshot sha256 27e3760e32ff…9014f · frozen 2026-10-01 · verify in console
Only Proctored

What the Browser-Extension Tools Can't Do

Most "AI governance for MSPs" products are a browser plug-in. Proctored is an agent on the device, which is the difference between watching one window and governing the machine.

Desktop and CLI

Content Inspection Beyond the Browser

The ChatGPT and Claude desktop apps, Claude Code, Codex and SDK traffic are inspected for content, not just allowed or blocked.

MCP governance

Govern the Tools Behind the Agents

Inventory the MCP servers configured in Claude Code, Claude Desktop, Cursor and VS Code, labelled by reach. Block remote servers; inspect every tool call to the approved ones.

Exceptions

A Block That Doesn't End in a Help-Desk Call

Time-boxed, device-bound, one-click approvals with a reason on record. Rules nobody should get around never admit them.

Proof

Coverage Evidence and Regression Alerts

Know which apps are really being inspected, and hear about it when a vendor's format change makes one go dark.

Beyond AI

Endpoint DLP and ZTNA in the Same Agent

USB, clipboard, data at rest, AirDrop and private app access, with the same rules and the same console. No second product.

Privacy

Redacted by Default

Only AI services are routed through the agent. Prompts are redacted in events unless a client chooses otherwise. Each device keeps its own certificate authority.

For MSPs

A New Recurring Line, Run From the Tools You Already Have

Per device, per month, billed like the rest of your stack. Three-tier tenancy, your brand and your help desk on every device and report, industry policy templates, and onboarding that takes an afternoon.

The Business Case

Security & privacy

Built to Be Trusted on the Endpoint

  • A device-unique CA whose key never leaves the machine
  • Signed releases verified before every update; macOS notarized by Apple
  • Tamper-resistant services that alert when enforcement is degraded
  • A single static binary, about 20 MB of memory, no third-party agents

What we read, what we keep, and what isn't done yet →

Questions MSPs ask

Straight Answers

Does it read everything my clients' staff do online?

No. The agent serves a proxy configuration that routes only AI-service domains, the web channels a client opts into, and decided MCP servers through itself. Every other site goes direct and is never seen. Prompt capture is redacted by default and can be turned off per client.

Does it slow the device down?

It's a single static binary using about 20 MB of memory. Inspection happens on the device for AI traffic only, so there's no round trip to a cloud and nothing to wait for on ordinary browsing.

Which AI tools does it cover?

One hundred and sixty-two catalogued apps across chat, coding, search, image, voice and meeting assistants, agents, legal and clinical AI, browser AI extensions, enterprise model APIs, embedded AI inside SaaS such as Notion AI and Microsoft 365 Copilot, plus local LLM runtimes, the desktop apps, command-line tools such as Claude Code and Codex, and MCP servers. Custom apps can be added by domain or client signature.

Can employees get around it?

On macOS a root daemon and on Windows a service running as LocalSystem own the proxy setting and re-apply it every thirty seconds. If enforcement is degraded the console alerts and your RMM health probe reports it. A standard user cannot remove the agent.

What happens when someone genuinely needs a blocked tool?

They ask for an exception from the block page or the device panel, with a reason and a duration. You get an alert, approve or deny in one click, and the grant is bound to that device and time-limited. Every message sent under it is still logged.

How is it priced?

Per device, per month, with partner pricing for service providers. There's no hardware and no setup fee. Ask us and we'll come back with numbers for your device count.

See it on a real device

A Demo Takes Thirty Minutes and Runs on a Real Machine, Not Slides

Tell us about your client base and we'll show Proctored catching a real block, approving a real exception, and producing a real report.

Request a DemoEverything It Does