Docs — Infser

Quick start

Four steps to a governed connection.

01 · Create your account

02 · Pair a device

Generate a pairing request from Infser and confirm it from your workspace.

03 · Review capabilities

The device reports the capabilities it actually supports. Infser does not assume access that was not reported.

04 · Use AI safely

Read-only actions can flow directly; higher-risk operations can require explicit approval before execution.

AI connection

Connect your AI client

ChatGPT, Claude, Gemini and Grok can use Infser as the governed MCP bridge. OAuth scopes and workspace permissions determine what each connection may request.

AI → InfserMCP

ChatGPT · Claude · Gemini · Grok → Infser

Execution path

Choose where the work runs

Use Browser Cloud for isolated ephemeral web automation when enabled, or a paired real device for machine and desktop actions.

cloudBrowser CloudChatGPT · Claude · Gemini · Grok → Infser → Browser Cloud

deviceReal deviceChatGPT · Claude · Gemini · Grok → Infser → Real device

Use your authorized computer or server for terminal, files, processes, services and supported desktop control. Example: “Open Chrome on my computer and change this setting.”

Agent install

Pair only devices you authorize

Generate a one-time pairing request, run the matching native installer and confirm the claimed device from your workspace.

Linuxshell

curl -fsSL https://infser.com/install-agent.sh | sudo sh -s -- --pair <TOKEN>

macOSshell

curl -fsSL https://infser.com/install-agent-macos.sh | sudo sh -s -- --pair <TOKEN>

WindowsPowerShell

irm https://infser.com/install-agent.ps1 -OutFile install-agent.ps1; .\install-agent.ps1 -Pair <TOKEN>

Install the native Infser agent only on systems you control. The signed release manifest selects the correct artifact; pairing still determines which workspace can use the machine.

The optional CLI adds diagnostic commands, but Node.js is not required by the native agent. @infser/cli · npx --yes https://infser.com/downloads/cli/0.3.0/infser-cli-0.3.0.tgz doctor

Browser Cloud

Open a site, click, type and complete web workflows in an isolated browser. The browser is separate from your local computer and is created ephemerally for the session.

Computer Core

Terminal, files, processes and services

Machine capabilities remain behind the same workspace, scope and policy boundary.

TerminalFilesProcessesServices

Desktop capabilities

Apps, windows, perception and input

Desktop capabilities are reported only when the local worker can provide them and remain subject to policy and approval.

Applicationsdesktop.apps.list · desktop.apps.open

Discover supported apps and open them using opaque application IDs.

Windowsdesktop.windows.list · desktop.windows.focus

List, focus, move, resize, minimize, maximize, restore and gracefully close windows without exposing native window handles.

Perceptiondesktop.state · desktop.inspect · desktop.screenshot

Desktop state returns topology, inspection returns a bounded accessibility tree and screenshot creates a temporary PNG Artifact.

Inputdesktop.click · desktop.type · desktop.key · desktop.scroll · desktop.drag

Click, type, key, scroll and drag use a recent state revision and opaque inspected element IDs; typed text stays transient.

Validation status

Permissions and validation stay explicit

Windows uses UI Automation with bounded SendInput fallback. macOS uses Accessibility with bounded CGEvent fallback and requires Screen Recording for screenshots. Linux uses X11 with AT-SPI, XTEST and native X11 capture; unsupported sessions do not advertise the capability.

Platform validation stays explicit

Linux X11 passed the real interactive v0.7 input gate for state, inspection, click, type, key, scroll, drag and stale-state rejection. Windows and macOS input still require authorized real-device E2E before cross-platform production-readiness is claimed.

From zero to connected