Product
Control what your teams send to AI
Sentlyx finds sensitive data in the prompts and files that employees send to AI apps. It acts on the laptop, before the data leaves, and shows all AI use in one console.
Four actions, for each type of data
- AllowLog the use for the console.
- WarnAsk the user for a reason.
- RedactSend placeholders such as [EMAIL_1]
- BlockStop the prompt, and say why.
How it works
How Sentlyx works
From the install on each laptop to one view of all AI use, in four steps.
Step 1
Deploy
Push the Sentlyx agent to every laptop through your MDM. The install is silent. If an agent stops, you get an alert.- Microsoft Intune Supported
- Configuration Manager (SCCM) Supported
4,812 devices protected Tamper alerts onStep 2
Detect on the laptop
Sentlyx finds keys, personal data, source code and your own code names in each prompt and each file upload. The scan runs on the laptop, so by default raw values stay there.Scan results · this laptop- CredentialAKIA••••••••3F7QAWS key
- Personal dataraj@varuna.inEmail
- Source codebilling/invoice.tsRepo match
- Custom termProject FalconCode name
Step 3
Apply your policy
For each type of data, select the action: allow, warn, redact or block. With redact, the AI sees only placeholders such as [EMAIL_1].
Start in monitor-only mode. Turn on the actions when you are ready.
Policy: Customer dataWarnedDraft a renewal note for Varuna Motors. Send it to raj@varuna.in.
This prompt contains a customer name and an email address.
Why do you need to send it?
The account lead approved this renewal note.
Step 4
See all AI use
One dashboard shows every AI app, every user and every incident. Send the events to your SIEM and keep your current alert flow.Time User App Rule Action 10:42 a.iyer ChatGPT Credentials Blocked 10:37 r.mehta Claude Customer data Warned 10:31 s.khan ChatGPT Source code Redacted Export to Splunk, Microsoft Sentinel, IBM QRadar or any webhook.
Detection
What Sentlyx finds
Two layers work together. Fixed detectors on the laptop catch the certain cases. Our own model, in your region, reads the masked prompt with your rules and decides the rest.
Credentials
API keys, cloud keys, private keys, tokens and passwords, with patterns and an entropy check.
Personal data
Email addresses, phone numbers, Aadhaar, PAN, SSN and card numbers, with checksums such as the Luhn check.
Source code
Code in a prompt, and code that matches the repositories of your company.
Your own terms
Project code names, customer names, keywords and fingerprints of your documents.
File uploads
The agent reads the text of a file before the upload and runs the same checks.
Prompt injection
The model is trained to ignore text in a prompt that tries to change its rules.
Rules in your own words
Write a rule as a sentence. The model reads your rules with each prompt, so a new rule works at once, with no training. Start in monitor-only mode, see what the rule finds, then turn on the action.
Do not share the pricing model of Project Falcon, or the names of its pilot customers.
Coverage
Works where your teams use AI
Sentlyx covers the AI apps that your employees use, in the browsers and on the systems that you already manage.
AI apps
- ChatGPT
- Microsoft 365 Copilot
- Gemini
- Claude
- Microsoft Copilot
- Perplexity
- DeepSeek
- Grok
- GitHub Copilot
- Meta AI
- Qwen
- Kimi
- Mistral Le Chat
Browsers
- Google Chrome
- Microsoft Edge
- Firefox
- Brave
Operating systems
- Windows
- LinuxNext
- macOSNext
By default, raw values never leave the laptop. If you turn on full-prompt capture, the laptop encrypts the prompt with your key before it sends it.
Read the security overviewDiscovery
Every AI app, also the unapproved ones
The agent sees which AI apps and websites each person uses, in every browser and in desktop apps such as ChatGPT and Claude. This works from the first day, with no traffic inspection.
Approve, leave unapproved or block
Set a status for each app. A block stops the whole app or AI domain on every managed device.
A catalog that we keep current
The list of AI apps, their domains and their desktop programs comes with each policy update.
Desktop apps and tools
The agent knows the desktop apps and command-line tools by their program, not only by the website.
| App | Users | Status |
|---|---|---|
| ChatGPT | 1,284 | Approved |
| Microsoft 365 Copilot | 932 | Approved |
| Claude | 411 | Approved |
| Perplexity | 187 | Unapproved |
| DeepSeek | 36 | Blocked |
The console
One place for all AI use
Admins sign in at console.sentlyx.com with single sign-on. Each role sees only what it needs.
Overview
Who uses AI, which apps, how many incidents and blocked prompts, and the trends by type of data.
AI apps
Each app with its users and its status: approved, unapproved or blocked.
Incidents
Each warning, redaction and block, with the masked sample and the reason that the user typed.
Policies
Your rules, the action for each type of data, and monitor-only mode while you tune them.
Users and devices
Each device with its agent version, its last check-in and its health, and each tamper alert.
Settings
Single sign-on, admins and roles, integrations, retention and the capture mode.
Deployment
Fits the tools that you already run
Sentlyx is built for large organizations, with the controls that a security review asks for.
Device management
A silent install on Windows with Microsoft Intune or Configuration Manager. Linux and macOS come next.
Single sign-on and SCIM
SAML or OIDC with Microsoft Entra ID, Okta or Google. SCIM keeps the users and groups in sync.
SIEM and webhooks
Send events and admin activity to Splunk, Microsoft Sentinel, IBM QRadar or a webhook.
Data residency
Your data stays in its home region. Today that is India; Europe and the Americas are planned.
Tamper alerts
An alert when an agent stops, when a part of it is turned off, or when it restarts unexpectedly.
Works offline
The agent keeps the last policy and the detectors on the laptop. You select what happens when the model is not available: allow or block.
Say yes to AI, safely.
Book a 30-minute demo and see Sentlyx on your own policies.