Book an Appointment
βData Loss Prevention for Generative AI

The DLP for ChatGPT, Claude & Copilot.

VamiGuard is the leading data-loss-prevention platform purpose-built for generative AI. Detect, redact, govern and audit every prompt — across browsers, desktop apps, IDEs and APIs, centrally managed, sovereignly hosted on Open Telekom Cloud.

ChatGPTClaudeMicrosoft CopilotGeminiGrokGitHub Copilot& any other LLM

Microsoft Edge · coming soon

< 0 msDetection latency per patternlocal · L1 · measured in Chrome 120
0 Bytesleave your browserno telemetry · no server calls
0 LLMscovered todayChatGPT · Claude · Copilot · Gemini · Grok · DeepSeek
Apache-2.0Open-source licensefully auditable

Figures refer to today's open-source extension — measured, not promised.

Live demo

See it in action.

VamiGuard intercepts the prompt as you type it — detecting emails, credit cards, passwords, API keys, JWTs and any custom regex — and replaces them with placeholders before the prompt ever leaves the browser. You stay in control: review every detection and decide on the redaction.

chat.openai.com
You're typing…
EMAIL detectedAPI KEY detectedPASSWORD detected

Live capture · 1:1 product

What you see in this clip

  1. 1A user pastes a real customer email into ChatGPT — full of PII and secrets.
  2. 2VamiGuard detects every email, phone number, credit card, password, JWT and API key in real time using local regex.
  3. 3Each value is replaced with a stable placeholder. The mapping lives in the page DOM and is reset on refresh.
  4. 4When the chatbot's response references the same placeholders, VamiGuard restores the original values so you can copy working code.

The category-defining DLP for generative AI

BSI C5ISO 27001GDPR Art. 25 / 32EU AI ActNIS-2Open Telekom Cloud
Coverage at a glance

Purpose-built for the GenAI tools your employees actually use.

ChatGPTOpenAI · Web + Desktop
ClaudeAnthropic · Web + Desktop + CLI
Microsoft CopilotM365 · Web + Add-In
GeminiGoogle · Web
GrokxAI · Web + X app
GitHub CopilotIDE · LSP plugin
Cursor · ContinueIDE assistants
Claude CodeCLI agent
Self-Hosted LLMsvLLM · Ollama · Mistral
Direct API CallsServer-to-server proxy
Future LLMsPolicy-based onboarding
Custom patternsAdd any domain
The four pillars of GenAI DLP

Detect. Redact. Govern. Audit. In every prompt.

VamiGuard delivers the four DLP capabilities every responsible GenAI deployment requires — from the first keystroke to the audit report.

01

Detect

Sensitive data is detected the moment it is typed — patterns, named entities, semantic context.

02

Redact

Tokens are replaced with placeholders before the prompt leaves the endpoint. The mapping stays local.

03

Govern

Central policies decide whether a finding is monitored, alerted, soft-blocked or hard-blocked — per group and channel.

04

Audit

Every decision is logged in pseudonymised form and is exportable to Splunk, Sentinel, QRadar or Elastic.

What VamiGuard recognises

Every category that should never reach an LLM.

Built-in patterns cover the data types your auditors ask about — custom patterns can be added freely. New patterns roll out without redeployment.

Personally identifiable information · PII

Identifying information

  • Email addresses
  • Phone numbers
  • Credit card numbers
  • IBAN
  • Social security numbers
  • Passport numbers
  • IP addresses
  • Names & addresses

Every pattern is switchable per group. Detected values receive a stable placeholder such as <EMAIL_1>, mapped consistently across the conversation.

Secrets · Tokens · Keys

Authentication material

  • AWS Access Keys
  • JWT Tokens
  • Bearer Tokens
  • GitHub Tokens
  • OpenAI API Keys
  • Stripe Keys
  • Passwords & secrets
  • High-entropy strings
  • Custom regex / patterns

Add plain-text terms (project codenames, internal product names) or full regex. Patterns can be imported and exported as YAML across teams.

Redact on send · Restore on receive

A second mapping layer between LLM response and user.

When Claude or ChatGPT returns a code snippet that references <TOKEN_1>, VamiGuard transparently substitutes the original value back — the user gets working code, the LLM never saw the value.

  1. 1User types: "Help me debug this auth header: Bearer sk-prod-abc…"
  2. 2VamiGuard redacts: "Help me debug this auth header: Bearer <TOKEN_1>"
  3. 3LLM responds with code that references <TOKEN_1>
  4. 4VamiGuard restores: the user sees working code with the original secret — ready to copy-paste
How detection works

Three layers. Each stricter than the last.

Most input is harmless and is cleared in milliseconds. Only truly ambiguous content reaches the deepest layer — low latency, high accuracy.

L1

Pattern · Regex · Entropy

Email, IBAN, credit cards, AWS keys, JWT, GitHub tokens, passwords. Runs entirely on device — the prompt never leaves it.

< 5 msOn Device
L2

Named entities · classifier

A local NER model recognises people, organisations, addresses, project names — augmented by your own word lists.

30 – 80 msOn Device
L3

LLM guardrail · semantics & intent

Only ambiguous prompts reach the EU-hosted guard model — already pre-redacted. Categories are configurable per tenant.

200 – 600 msEU Region
Policy modes

Four levels of decision.

Every detection rule binds to one of four response modes. Different groups, channels and times of day can run different modes simultaneously.

M1

Monitor

Silent observation. Events are logged for reporting. Ideal for the learning phase.

M2

Alert

Visible warning. The user can still submit. Optionally with a notification to manager or SOC.

M3

Soft-Block

Submit is paused. The user enters a written justification — auditable and retrievable.

M4

Hard-Block

The submit path is removed. Three-layer enforcement via MDM, network and Conditional Access.

Coming soon — centralized management is on the roadmap. Today's open-source extension shows warnings directly to the user.

Coverage

Every channel through which your users reach an LLM.

Browser alone is not enough. VamiGuard covers desktop apps, IDEs, mobile devices and server calls — all governed by a single central policy.

Browser

Chrome · Edge · Firefox

Desktop apps

Claude · ChatGPT · Copilot

IDE plugins

VS Code · JetBrains · Cursor

Mobile (MAM)

iOS · Android · Intune · Jamf

API Gateway

OpenAI · Anthropic · Server-to-Server
Sovereignty & compliance

Your data — in your jurisdiction.

VamiGuard Cloud runs exclusively in the EU on Open Telekom Cloud — operated by T-Systems under German law. Or fully on your own infrastructure.

Three availability zones in the region eu-de. KMS-backed encryption with optional BYOK in your own HSM. No US parent company. No CLOUD Act exposure. Full audit-trail ownership in the on-premises tier.

BSI C5Type-2 attested
ISO 27001+ 27017 / 27018
GDPRArt. 32 / 25 / 5
EU CCoCCode of Conduct L3
EU AI ActArt. 9 documented
NIS-2Incident reporting
TCDP 1.0Level 3
BYOK / HSMFIPS 140-2 L3

Coming soon — the points above describe the targeted compliance posture for the upcoming cloud and sovereign plans, not the current open-source version.

Plans

Open source as the foundation. Enterprise on top.

The browser extension is and remains free, open source and Apache-2.0. Central management is added on top whenever you are ready — data sovereignty stays with you in any case.

Community

€ 0Self-managed

Browser extension installed by the user. Local patterns, local mapping, no telemetry, Apache-2.0 licensed.

  • PII redactor
  • Custom regex & patterns
  • Fully client-side
  • Apache-2.0
Install for free

Sovereign

CustomContact Sales

Helm chart for your own data centre or OTC tenancy. Air-gap-capable. Bring your own LLM and HSM. Audit-trail ownership.

  • Everything in Cloud
  • Helm Chart (CCE / OpenShift)
  • Bring-your-own LLM
  • HSM & BYOK
  • Air-gap update pipeline
  • White-label option
Contact Sales
Get started

Install it in a minute. Manage it in a quarter.

Try the free open-source extension on Chrome today. Reach out to us as soon as you want to bring central management to your organisation.

Protect Your Organization Now!

Contact us for an individual consultation and security solution tailored to your requirements.

Valeri Milke, CEO of VamiSec

"Only when all instruments are well-tuned does your organization become secure and compliant."