PromptGlass documentation

PromptGlass shows you how your organization uses AI. It records which employees and teams use which AI tools, captures prompts and responses, catches sensitive data before it leaves, and reports on adoption, efficiency and spend. It runs as SaaS, as a self-hosted deployment on Kubernetes, or fully air-gapped on your own hardware.

These docs cover deployment, the four capture methods (we call them sensors), SaaS connectors, detection, policy, analytics, privacy controls and integrations.

Start here

QuickstartFrom SSO to your first AI usage dashboard in about 15 minutes.

ArchitectureSensors, gateway, detection, storage and how data flows between them.

Deploy on KubernetesSelf-host PromptGlass with the official Helm chart.

Policy referenceEvery key in a PromptGlass policy file.

Capture AI usage

Sensors

Gateway forward proxy · Gateway API mode · Browser extension · Endpoint agent · Chaining behind your SWG

SaaS connectors

Microsoft 365 Copilot · OpenAI · Anthropic · GitHub Copilot · Google Gemini · AWS Bedrock · Azure OpenAI

Protect and govern

Built-in detectorsPII, PHI, PCI, secrets, source code and confidential business terms.

PoliciesAllow, log, coach, redact, block or route to an approved model.

Employee privacyPseudonymization, four-eyes re-identification and works-council mode.

Data residency and retentionRegions, storage modes and retention by data class.

Understand usage

AdoptionActive users, teams and apps over time.

EfficiencyPrompt quality, re-prompt rate and top use cases.

Cost optimizationToken spend, idle seats and model right-sizing.

Shadow AIDiscover unsanctioned AI apps and score their risk.

Connect your stack

SIEMSplunk, Microsoft Sentinel, Elastic, QRadar and Chronicle.

SSO and SCIMEntra ID, Okta and Google Workspace.

MCP serverAsk Claude or Copilot about your organization's AI usage.

REST APIQuery events, manage policies and export data.

New to the product? Read the platform overview or see how it works.