Four sensors in,
one decision out.
PromptGlass captures AI usage wherever it happens: in the browser, on the endpoint, on the network and in SaaS admin APIs. Every interaction is identified, inspected and decided in real time, then analyzed.
The pipeline
Capture → Identify → Detect → Decide → Analyze
- 1
Capture
A sensor observes an AI interaction: a prompt, upload, API call or audit event.
- 2
Identify
The interaction is joined to a person, team and cost center from your identity provider, then pseudonymized for analysts.
- 3
Detect
Detectors scan content for personal, health and payment data, secrets, source code and your custom terms.
- 4
Decide
The policy engine picks an action: allow, log, coach, redact, block or reroute, in milliseconds.
- 5
Analyze
Events land in the analytics store for adoption, efficiency, cost and risk reporting, and stream to your SIEM.
Sensors
Use one, or use all four
Forward proxy + AI API gateway
PromptGlass Gateway
Inspects only traffic to domains in the AI catalog, configured by PAC file, explicit proxy or chaining from your existing secure web gateway. For software, it exposes OpenAI- and Anthropic-compatible endpoints, so apps switch by changing a base URL.
export OPENAI_BASE_URL=https://gateway.corp.internal/v1Setup guide →- ✓TLS inspection limited to AI domains
- ✓Streaming, tools and batch supported
- ✓Virtual keys, budgets and model routing
- ✓Stateless pods that scale horizontally
Chrome · Edge · Firefox · Safari
Browser extension
Sees prompts, pastes and uploads inside web AI apps, including which account is signed in, and shows coaching messages in the page. Deployed silently through your device management.
Setup guide →- ✓No network change
- ✓Personal vs. corporate account detection
- ✓Inline coaching and redaction
- ✓Managed via Intune, Jamf, Google Admin or GPO
Windows · macOS · Linux
Endpoint agent
Discovers AI desktop apps, IDE assistants and local model runtimes, and attributes them to the signed-in user. Covers what no network sensor can see.
Setup guide →- ✓Desktop AI apps and IDE plugins
- ✓Local LLMs (Ollama, LM Studio)
- ✓Installed AI browser extensions
- ✓Low-overhead, signed binaries
Read-only, no interception
SaaS API connectors
Pull usage, audit and seat data from the AI services you already license, through their official admin APIs.
Setup guide →- ✓Microsoft 365 Copilot via Microsoft Graph
- ✓ChatGPT Enterprise and Anthropic admin APIs
- ✓GitHub Copilot metrics
- ✓Bedrock, Azure OpenAI and Vertex AI
Frequently asked questions
Do we need all four sensors?
+
No. Most organizations start with the browser extension and one or two SaaS connectors, then add the gateway for API traffic and the endpoint agent for desktop and local AI.
Does the gateway decrypt all of our HTTPS traffic?
+
No. It only inspects connections to domains in the AI catalog, and tunnels everything else without decryption. You can also chain it behind an existing secure web gateway.
What happens if PromptGlass is unavailable?
+
Each policy sets fail-open or fail-closed behavior. Gateways run as multiple replicas behind your load balancer, and the browser extension keeps applying the last known policy.
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.