Your team is already pasting customer data into ChatGPT and Claude. Nobody means any harm. An engineer drops a stack trace into a prompt to debug it faster, and a customer record goes with it. The leak isn't an attack. It's a Tuesday.
SUPERWISE Sentinel stops that data on the machine, before it reaches the model. Version 1.38.0, released September 15, 2026, makes Sentinel the center of the platform and gives admins much more control over how it behaves.
What's New in 1.38.0
- Sentinel is now front and center. Sentinel is the default product when you log in. A new home screen shows what it's doing for you, with interactions, mitigated issues and latency at a glance, plus quick actions. The new product switcher takes you to the rest of the platform whenever you need it.
- Admin and read-only roles. Admins manage Sentinels, guardrail policies, token optimization and organization settings. Read-only Viewers keep full visibility without edit access.
- Editable guardrails for each Sentinel. Admins can tailor every Sentinel's guardrails. Choose whether a guardrail blocks a request or remediates it, with your own error message or replacement text. Pick exactly which categories of personal data to scan for, and drag guardrails to reorder the pipeline.
- Token optimization (MVP). Turn on Headroom compression to shrink prompts before they reach the model, cutting token usage, cost and latency. It's off by default, and an admin turns it on for each Sentinel.
- New plans. Solo, Pro and Business tiers, with limits based on your Sentinel usage. See pricing.
Read the full 1.38.0 release notes.
Recently Shipped
If you haven't looked at Sentinel in a while, these landed in the two releases before this one:
- FinOps dashboard (1.37.0). See token use and efficiency across your fleet: total tokens, cache-hit rate, reasoning ratio and output-to-input ratio, broken down by provider, model and Sentinel. It uses metadata only, refreshes live, and exports to PDF, PNG or CSV.
- Guardrails managed per gateway (1.37.0). Add or remove Secret detection, PII redaction, and Rudeness & toxicity guardrails on any Sentinel, right from the app. Changes take effect on the gateway's next heartbeat.
- Broader, stricter PII redaction (1.37.0). The PII guardrail now covers passports, driver's licenses, tax and national IDs, IBANs and bank accounts, IP and MAC addresses, VINs, dates of birth and medical record numbers. It also matches labeled IDs more precisely, which cuts false positives.
- Activity Log (1.36.0). A live, fleet-wide stream of every interaction passing through your gateways, showing provider, model, outcome and latency as it happens. Metadata only: your prompts and responses never leave your environment.
What Sentinel Does to a Prompt
Sentinel is a gateway that runs on your machine, between your tools and the model. It inspects every request, replaces secrets and personal data before anything leaves, and records every check. Here's a prompt as someone typed it, and what the model actually received:
What was typed What the model received
Here's the failing record: Here's the failing record:
name Dana Whitfield name <NAME>
email d.whitfield@acme.com email <EMAIL>
key sk-proj-abc123 key <API_KEY>The person keeps working. Nothing changes for them except what leaves the machine.
Where Sentinel Works
| What you use | Windows | Mac & Linux | Chrome / Edge |
|---|---|---|---|
| Claude (Claude Code and claude.ai) | Protected | Protected | Protected |
| OpenAI (Codex and chatgpt.com) | Protected | Protected | Protected |
| Your own apps and agents | Protected | Protected | — |
| Gemini, Copilot, Grok, Perplexity, DeepSeek | Coming soon | Coming soon | Coming soon |
Known Limitations
- Windows code signing is being finalized. Until it's done, Windows SmartScreen shows a warning when you run the installer. Choose More info, then Run anyway.
- The browser extension is Windows only for now.
- Gemini, Copilot, Grok, Perplexity and DeepSeek support is in development.
Get Started
Windows
Download Sentinel for Windows and run the installer. If SmartScreen appears, choose More info, then Run anyway. Installing for every user on a computer? Use the per-machine installer.
The Windows installation guide covers setup, including Claude Code and claude.ai.
Mac and Linux
Three commands and you're protected:
# 1. Install the CLI (Python 3.11+)
pip install superwise-sentinel-cli
# 2. Deploy a gateway
sentinel quickstart
# 3. Prove it redacts, with no model call
sentinel test --text "email john.doe@example.com · key sk-proj-abc123"You should see:
✓ redacted: email <EMAIL> · key <API_KEY>Docker Desktop needs to be running. The Mac and Linux installation guide has the full walkthrough.
Your Own Apps and Agents
To send your own applications' AI traffic through Sentinel, see instrumenting your AI traffic. For Codex and similar tools, see auto-configuring apps.
Try It Without Installing
Run a real prompt through a live gateway. No install, no sign-up.
Pricing
Plans start at $10 per month, and you can try Sentinel free for 30 days. See pricing.
Rolling Sentinel out across your team? Email us and we'll help you deploy it for your business.
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.