Skip to content
FREE30-day AI usage assessment→

See how your company really uses AI.Every prompt. Every app. Every dollar.

PromptGlass monitors AI usage by employee across ChatGPT, Copilot, Claude, Gemini, coding assistants and model APIs. It captures prompts, stops sensitive data before it leaks, and shows where AI saves time and where it wastes money.

Start free assessmentBook a demo
  • ✓Cloud or on-prem
  • ✓Kubernetes-native
  • ✓Privacy by design

4 sensors

proxy, browser, endpoint, API

120+

built-in sensitive-data detectors

<15 ms

p99 inline inspection overhead

3 ways

to deploy: SaaS, Kubernetes, air-gapped

Visibility into the AI tools your people already use, approved or not

The AI visibility gap

AI adoption is outrunning AI governance.

Employees adopted generative AI faster than any technology before it. Most organizations still can’t answer three basic questions.

Shadow AI you can’t see

Personal ChatGPT accounts, AI browser extensions, coding assistants on expense reports and local models on laptops. Surveys miss most of it.

Data leaving one prompt at a time

Customer lists, contracts, patient notes and API keys are pasted into AI tools every day, usually by people just trying to get work done.

AI spend without a ledger

Per-seat copilots nobody opens, duplicate assistants, and API bills no team owns. Finance sees the invoice, not the value.

One platform

See it. Protect it. Optimize it.

PromptGlass covers the full lifecycle of workplace AI, from discovery to enforcement to ROI, in a single console.

See

Every AI app, user, prompt and dollar.

AI usage monitoringWho uses which AI tool, how often, for what, at what cost.Prompt & response captureSearchable record of prompts, responses and uploads, under your privacy rules.Shadow AI discoveryFind unapproved AI apps, personal accounts, extensions and local models.

Protect

Stop data leakage without stopping work.

AI data loss preventionDetect PII, PHI, secrets and code in prompts. Coach, redact or block inline.AI policy enforcementAllow, coach, redact, block or reroute, per app, group and data class.AI gateway & proxyOpenAI/Anthropic-compatible proxy with keys, budgets, routing and DLP.

Optimize

Spend less and get more from AI.

AI cost managementSpend by team and model, idle seats, duplicate tools, right-sizing.AI productivity & efficiencyUse cases, prompt quality, rework rates, and where training pays off.

How it works

Four sensors. One timeline per person.

Use one sensor or all four. Each AI interaction is tied to a person, a team and a cost center from your identity provider, then inspected, decided and analyzed in real time.

  • 01

    Gateway

    Forward proxy for AI domains only, plus an OpenAI- and Anthropic-compatible API endpoint for apps and agents.

  • 02

    Browser extension

    Chrome, Edge, Firefox and Safari. Pushed by Intune, Jamf or Google Admin. No network changes.

  • 03

    Endpoint agent

    Finds desktop AI apps, IDE assistants and local LLMs that never touch the network.

  • 04

    API connectors

    Read-only links to Microsoft 365 Copilot, ChatGPT Enterprise, GitHub Copilot, Bedrock and more.

See the architecture →
AI data loss prevention

Protect data without blocking work.

Blocking AI pushes people to their phones. PromptGlass redacts the sensitive parts, lets the prompt through, and restores them in the answer. People keep working; the model never sees the secret.

What the employee typed

Draft a renewal email to Dana Whitfield at Northwind Health. Her account 4111 1111 1111 1111 renews on the 30th.

What the model received

Draft a renewal email to [PERSON_1] at [CUSTOMER_1]. Her account [CARD_1] renews on the 30th.

What the employee got back

“Hi Dana, I hope all is well at Northwind Health. A quick note that your plan renews on the 30th…” (names restored, card number never returned)

policies/customer-data.yaml
name: customer-data-in-ai
scope:
  apps: ["*"]
  groups: ["All employees"]
rules:
  - match: { detector: pci.card_number }
    action: redact
    restore: false
  - match: { detector: [pii.person, edm.customers] }
    action: redact
    restore: true
  - match: { account: personal, upload: true }
    action: block
    coach: "Use the company workspace for files."
on_inspection_error: fail_open
allowlogcoachredactblockreroute

Explore AI data loss prevention →

Answers for every leader

One source of truth for how AI is used.

Security, IT, finance and legal look at the same data and answer different questions.

CISO & security

“Where is sensitive data going?”

Leakage events by data class, shadow AI risk scores, SIEM-ready alerts and enforcement that doesn’t block work.

CIO & IT

“Which AI tools should we standardize on?”

Adoption by app and team, overlap between tools, and which approved tools people actually prefer.

CFO & finance

“What are we getting for our AI spend?”

Seats vs. active users, token spend by cost center, and a reclaim list worth real money.

Legal, privacy & HR

“Can we prove responsible AI use?”

AI inventory, policy evidence for audits, and employee-privacy controls designed with works councils in mind.

Deploy your way

Your cloud, our cloud, or no cloud at all.

The same platform runs as managed SaaS, in your own Kubernetes cluster, or fully disconnected. Prompt content can stay inside your perimeter.

Fastest

PromptGlass Cloud

Fully managed SaaS in US, EU or Canada regions. Live in a day.

Most popular

Self-hosted Kubernetes

Helm chart for EKS, AKS, GKE, OpenShift or bare metal. Prompts never leave your cluster.

Sovereign

Air-gapped on-prem

Signed images and an offline AI catalog for classified and regulated networks.

$helm install promptglass oci://registry.promptglass.ai/charts/promptglass -n promptglass --create-namespace

Compare deployment options →

Industries

Built for regulated work.

Detectors, policies and reports tuned for the data each industry must protect.

Healthcare→Financial services→Legal→Government & public sector→Software & technology→Manufacturing→

Compliance

Evidence, not just policy.

Inventory, monitoring records and enforced controls that support the frameworks auditors ask about.

EU AI Act→ISO 42001→NIST AI RMF→GDPR→HIPAA→SOC 2→PCI DSS→

Frequently asked questions

What is PromptGlass?

+

PromptGlass is an AI usage monitoring and governance platform. It shows which employees use which AI tools, captures prompts and responses under your privacy rules, detects and stops data leakage, and measures AI adoption, efficiency and cost. It runs as SaaS or self-hosted on Kubernetes.

How does PromptGlass capture AI usage?

+

Through four sensors that you can combine: an AI-aware gateway (forward proxy and OpenAI/Anthropic-compatible API proxy), a browser extension, an endpoint agent, and read-only API connectors for services such as Microsoft 365 Copilot, ChatGPT Enterprise and GitHub Copilot.

Can PromptGlass run on-premises?

+

Yes. PromptGlass is available as PromptGlass Cloud (US, EU or Canada regions), self-hosted on any Kubernetes cluster with a Helm chart, or fully air-gapped with an offline AI catalog.

Does PromptGlass inspect all of our web traffic?

+

No. The gateway only inspects traffic to domains in the AI catalog and tunnels everything else untouched. The browser extension needs no network change at all.

How does PromptGlass protect employee privacy?

+

Users are pseudonymized for analysts by default, content storage can be full, redacted, metadata-only or hashed per policy, and revealing an identity requires a second approver. Every access is audited.

Which AI tools does PromptGlass monitor?

+

ChatGPT, Claude, Microsoft 365 Copilot, GitHub Copilot, Gemini, Cursor, Perplexity, DeepSeek and many more, plus model APIs (OpenAI, Anthropic, Azure OpenAI, Amazon Bedrock) and local LLMs such as Ollama.

How long does deployment take?

+

A SaaS rollout with SSO, the browser extension and one connector usually takes a day. A self-hosted Kubernetes deployment typically takes a few days including network and identity integration.

Find out how your company really uses AI.

Start a free 30-day AI usage assessment: a complete shadow AI inventory, the sensitive data leaving, and the AI seats you are paying for but not using.

Start the free assessmentBook a demo