How It Works
Waxell Endpoints has three ways of seeing AI use on a machine, all fed by one policy and all reporting to your workspace:
┌──────────────────────────── Managed device ─────────────────────────────┐
│ │
│ Browser (Chrome / Edge / Firefox) Coding agents & AI apps │
│ ┌──────────────────────────┐ (Claude Code, Codex, Cursor, │
│ │ Waxell AI Monitor │ Gemini CLI, Claude Desktop) │
│ │ extension │ │ │
│ │ · reads the AI chat │ │ managed hooks │
│ │ · enforces browser DLP │ ▼ (before each │
│ └────────────┬─────────────┘ ┌──────────────────┐ action) │
│ native messaging │ wax-hook │ │
│ ▼ │ allow / ask / │ │
│ ┌──────────────────────────────────────┐ │ deny │ │
│ │ Waxell agent │◄┴──────────────────┘ │
│ │ · policy daemon (pulls the Guard) │ │
│ │ · redacts secrets on the device │ App scanner + network │
│ │ · uploads captures & decisions │◄─ monitor (which AI apps, │
│ │ · health beacon every 60 s │ which hosts, which user) │
│ └──────────────────┬───────────────────┘ │
└──────────────────────┼──────────────────────────────────────────────────┘
▼ HTTPS
Your Waxell workspace
(Device Activity · Captures · Decisions · Alerts · Health · Guard)
The pieces on the device
- The Waxell agent. A signed service (macOS app with system extensions; Windows service) that enrolls the device, pulls policy, redacts and uploads captures, and reports health.
- The browser extension. Waxell AI Monitor, force-installed into Chrome, Edge and Firefox. It reads AI conversations in the page and applies browser policy before a prompt is sent. See Browser Extension.
- Managed hooks. The agent registers Waxell's hook in each coding agent's administrator-owned configuration. The tool calls it before running a command, touching a file or submitting a prompt. See Coding Agents (Hooks).
- App scanner and network monitor. Inventory every AI app installed on the device, and record which app reached which AI host, when, and as which user — metadata only, nothing decrypted.
- Watchers. Read transcripts that some tools keep locally (and, when you turn it on, the ChatGPT desktop app's conversation) for after-the-fact visibility.
Enrollment
On first run the device enrolls with your workspace using the key in its configuration profile (or the one you pasted in a manual install), and receives its own device identity. From then on every capture, decision and health report is attributed to that device and, where known, its owner.
The owner is the device's user. Set it on the device's page, or let your MDM fill it in — Waxell can read device owners from Hexnode, and resolve aliases across your directory (Google Workspace, JumpCloud).
The policy loop
Policy is the Guard, a cascade of layers from the whole workspace down to a single device. Each device checks for a new effective policy about once a minute. When it changes, the agent writes it locally and pushes it to the extension and the hooks, which enforce it from that local copy. That is why enforcement keeps working when a laptop is offline or Waxell is unreachable.
Redaction on the device
Before any capture leaves the machine, the agent redacts secrets in it: API keys, bearer tokens, cloud and SaaS credentials, private keys, JWTs, connection strings, card numbers (checksum-validated), SSNs and labelled secrets. A capture that had something removed is marked redacted; Waxell never receives the original. You can add your own detectors. See Privacy & DLP.
Redaction protects Waxell's copy. Whether the user's prompt is allowed to reach the AI service at all is a separate, enforcement decision — browser DLP mode and agent DLP mode.
What Waxell does not do
- It does not decrypt network traffic. Browser AI is read in the page, and coding agents report through their own hooks. (TLS interception existed in earlier versions and is disabled in the current agents.)
- It does not capture non-AI browsing.
- It does not need the user to do anything once deployed through MDM.
What you see in the console
Everything is under Endpoints in your workspace:
- Overview — AI usage and policy across the fleet.
- Activity — pick a device and see every AI tool on it, session by session, turn by turn, with the commands, files and decisions in line.
- Captures — every captured browser conversation, redacted, filterable by device and account type.
- Alerts — live conditions, behavioural anomalies and coverage gaps in one list.
- Inventory — discovered AI apps, the Shadow AI report and the AI catalog.
- Guard and Decisions — the policy, and every allow, ask and deny.
- Devices, Health, Site Adapters, Agents — the fleet.
- Setup — download installers and profiles.
Next: Key concepts.