Inspect the prompt
Every prompt from a covered tool passes through Sentinel first.
SUPERWISE® Sentinel
Your team is pasting customer data into ChatGPT and Claude.
A ban does not stop AI. It hides it. Keep the tool, protect the customer.
Your prompt
Sentinel caught:
What the model receives
Draft a short apology about the late refund: Email: {{REDACTED}} Phone: {{REDACTED}} Card: {{REDACTED}}
Model response
Three examples of what Sentinel does to a prompt. A customer record: the email d.whitfield@acme.com, the phone number (415) 555-0142, and the card number 4111 1111 1111 1111 are replaced with {{REDACTED}} before the prompt reaches ChatGPT, and the model still writes the reply. An API key in a prompt: the AWS access key and the alert email address are replaced before the request leaves the computer. A toxic reply: a request to insult a customer has the toxic language replaced before it reaches the model.
Typing in ChatGPTSee it work in 30 seconds →
AI firewall
What an AI firewall does
An AI firewall sits between your team and the AI model and checks each prompt on its way out.
Redact or block
Sentinel catches sensitive values in a prompt before it leaves, then redacts or blocks them, whichever the admin chooses. The admin sets the replacement text.
Record the check
Every check is recorded as metadata: when it ran, which guardrails ran, and the outcome. Prompt content stays private.
In plain English
AI DLP that runs on the desktop
Sentinel checks covered prompts before they leave the computer and replaces the sensitive values it recognizes. Prompts go straight to the AI provider. SUPERWISE receives only check metadata: counts, which rule fired, and latency.
01It reads what is about to be sent
Every prompt from a covered tool passes through Sentinel first.
02It swaps out anything sensitive
It replaces the sensitive values and secrets it recognizes before the request leaves.
03It keeps a record you can show
The log shows what Sentinel replaced before each request left the PC.
What it catches
What Sentinel catches
Sentinel replaces email addresses, phone numbers, Social Security numbers in the 123-45-6789 format, credit card numbers, IBANs, IP and MAC addresses, full-length OpenAI keys, and AWS secret keys when labeled. Bank account numbers, tax IDs and EINs, national IDs, VINs, dates of birth, and medical record numbers are replaced only in the labeled formats the gateway recognizes.
Custom patterns, such as your own policy numbers or customer IDs, are on the roadmap.
Where Sentinel works
The tools your team already uses.
The tools marked Supported or Preview below are covered wherever the machine runs.
On the machine
Terminal agents and the apps you build, covered wherever they run.
| Tool | Windows | Mac & Linux |
|---|---|---|
| Claude Desktopsetup guide Anthropic's desktop app. On Windows, it requires an Anthropic API key. On macOS and Linux, it requires the Anthropic provider and a one-time browser sign-in. | Supported | Supported setup guide |
| Claude Codesetup guide Anthropic's terminal agent | Supported | Supported setup guide |
| Codex CLIsetup guide Codex CLI: Supported on Windows, Mac and Linux when the tool uses an API key. | Supported* | Supported* |
| Your own apps and agentssetup guide OpenClaw, LangChain, or any framework | Supported | Supported |
| Antigravity (agy)Google's coding agent | Supported* | Supported* |
*When the tool uses an API key. Signing in with a ChatGPT or Google account is not covered yet.
In the browser
Preview extension
The browser extension is in preview for Chrome and Edge on Windows.
| Chat site | Status |
|---|---|
| Claudesetup guide claude.ai | Preview |
| ChatGPTchatgpt.com | Preview |
| Gemini, Copilot, Grok, Perplexity, DeepSeekIn development | Coming soon |
A Mac desktop app is coming soon. The Mac and Linux CLI stays available today.
SupportedPreviewComing soon
Using something not listed? Tell us what to cover next
What you see
Every Sentinel, and what it stopped
The record
- Each Sentinel event identifies the Sentinel that handled it (the protected computer or gateway, by its name), not the person using it.
- Kept for one year while the customer's account is active.
- Viewable and reportable in the SUPERWISE app today; export to a SIEM is on the roadmap.
Price and start
Free for 30 days. Then choose the plan that fits.
Self-serve plans are $10, $25 or $50 a month. See plans. Sentinel for Business. Sentinel for IT.
Free for 30 days, no card. Then $10, $25, or $50 a month, with 1 to 5 Sentinels per plan. Buy online. No sales call. See plans.
Deploy it your way
One Sentinel or one for every team
Windows app
Install it for one person without admin rights.
CLI
Run Sentinel from the terminal on Windows, macOS or Linux.
Team gateway
Run it in your own cloud for a team. Read about the LLM gateway or use the gateway deployment guide.
Under the hood
For engineers: how the AI gateway runs
Sentinel is a proxy between your tools and the model. It runs on the machine or in your own cloud, and requests go straight to the provider. Only the interaction log’s metadata syncs to the SUPERWISE platform.
- Keep secrets out of Claude Code
- How the LLM gateway handles personal data
- PII redaction guide
- Guardrails across the SUPERWISE platform
- How SUPERWISE handles data: trust center
routeTraffic points at a local gateway
Claude Code, Codex, your own agents and the browser extension send requests to the gateway instead of straight to the provider. No SDK, no wrapper, no code change.
inspectGuardrails run before the request leaves
Guardrails check the prompt on its way out. The model's response comes back unchanged. The admin chooses whether Sentinel redacts or blocks, and sets the replacement text.
forwardForwarded, then logged
The cleaned request goes on to OpenAI, Anthropic, Google Gemini, and any OpenAI-compatible endpoint added as a custom provider. The check, and which rule fired, shows in Live activity.
Questions buyers ask
Keep reading
How Sentinel compares: Harmonic Security, Microsoft Purview, and Prompt Security.
AI gateway questions
What buyers ask about running an AI gateway inside their own environment
An AI data loss prevention tool checks prompts before they reach a model and applies a policy at that point. SUPERWISE Sentinel replaces the supported identifiers and credentials in requests to covered ChatGPT and Claude routes, then for each request Sentinel records metadata only: counts, which rule fired, and latency. Prompts go straight to the AI provider; SUPERWISE receives only check metadata.
Use an approved account, a written data rule, and an enforcement point on each covered route. SUPERWISE Sentinel checks supported ChatGPT requests before they leave the device, replaces the supported customer identifiers with placeholders, and for each request Sentinel records metadata only: counts, which rule fired, and latency. Prompts go straight to the AI provider; SUPERWISE receives only check metadata.
Sentinel controls the AI routes it covers and records metadata about each check, which gives administrators evidence of AI use on those routes. Pair it with network, browser, and endpoint discovery to find every unapproved tool.
Sentinel is a small app that checks what your team sends to AI tools like ChatGPT and Claude. Sentinel catches sensitive values such as email addresses, phone numbers, card numbers, and API keys before a prompt leaves. The admin chooses whether Sentinel redacts or blocks, and sets the replacement text. Prompts go straight to the AI provider. SUPERWISE receives only check metadata: counts, which rule fired, and latency.
An AI gateway sits between your applications and the large language models they call, inspecting and controlling every request. SUPERWISE Sentinel runs as an AI gateway inside your own environment, so prompts are checked before they leave your network instead of after.
It is a proxy, but one you run. Sentinel is a gateway on your own computer or in your own cloud account. Your AI tools send their requests to it; it checks each prompt against your policy and forwards it to the provider the tool already uses. Requests go straight from your gateway to the provider. Only configuration and metadata (counts, which rule fired, latency) sync with the SUPERWISE control plane.
OpenAI, Anthropic, Google Gemini, and any OpenAI-compatible endpoint added as a custom provider.
Yes. Claude Desktop is supported through the Sentinel CLI on macOS and Linux. On macOS and Linux, it requires the Anthropic provider and a one-time browser sign-in.
Every check is recorded as metadata: when it ran, which guardrails ran, and the outcome. Prompt content stays private.
You wouldn’t put a network online without a firewall.
Your team is already using AI. Sentinel is the layer that watches what it sends.
Free for 30 days, no card. Then $10, $25, or $50 a month, with 1 to 5 Sentinels per plan. Buy online. No sales call. See plans.