Skip to content
PromptGlassDocs

Endpoint agent

The PromptGlass endpoint agent finds AI that the browser can’t see: desktop apps (ChatGPT desktop, Claude Desktop, Copilot), AI coding assistants inside IDEs (GitHub Copilot, Cursor, Continue, Codeium/Windsurf, JetBrains AI), command-line agents, and local model runtimes like Ollama and LM Studio. It reports inventory and usage, and it can steer devices to the gateway.

What it does

Section titled “What it does”
  • Discovery: an inventory of installed and running AI apps, IDE extensions, and model files over a size threshold (*.gguf, *.safetensors).
  • Usage telemetry: active time per AI app, process launches, and network destinations matched against the AI catalog.
  • Local model monitoring: detects Ollama or LM Studio servers listening on localhost and, if enabled, records model names and request counts through their local APIs.
  • Proxy steering: sets per-app proxy settings, or installs the gateway CA, so desktop AI traffic reaches the forward proxy.
  • Health: reports whether the browser extension is installed and enabled.

The agent is observe-only. Enforcement happens at the gateway or in the extension.

Install

Section titled “Install”
OS Package Deploy with
Windows 10/11, Server 2019+ PromptGlassAgent-x64.msi Intune (Win32 app), SCCM, GPO
macOS 13+ (Apple silicon & Intel) PromptGlassAgent.pkg + PPPC profile Jamf, Kandji, Intune
Linux (glibc 2.31+) .deb, .rpm Ansible, Puppet, your package repo
Terminal window
msiexec /i PromptGlassAgent-x64.msi /qn TENANT=acme.eu ENROLLMENT_TOKEN=pgt_xxx
Terminal window
sudo installer -pkg PromptGlassAgent.pkg -target /
sudo /Library/PromptGlass/bin/pgagent enroll --tenant acme.eu --token pgt_xxx
Terminal window
sudo apt install ./promptglass-agent_2026.11.0_amd64.deb
sudo pgagent enroll --tenant acme.eu --token pgt_xxx

Configuration

Section titled “Configuration”

Agents pull configuration from the console. You can override it locally for testing:

/etc/promptglass/agent.yaml
discovery:
  interval: 6h
  modelFiles: { enabled: true, minSizeMB: 500, paths: ["~", "/opt"] }
localModels:
  ollama: { enabled: true, port: 11434, recordPrompts: false }
proxySteering:
  enabled: true
  apps: [chatgpt-desktop, claude-desktop, cursor]

Resource usage

Section titled “Resource usage”

The agent is a single native binary. It is idle between discovery scans and reports events in batches. Check it with:

Terminal window
pgagent status
pgagent diagnose --upload   # sends a support bundle to your tenant