agent-desktop
Agent Desktop gives any agent reliable computer use on the desktop. Built with Rust, it sees any app's real UI structure through OS accessibility trees and oper
Sign in to rate this tool.
Overview
Agent Desktop, marketed as agent-browser, is a command‑line tool that lets AI agents control a web browser. It runs as a native Rust binary and communicates with Chrome through the Chrome DevTools Protocol. The focus is on delivering compact, text‑based output that fits within the limited context windows of large language models.
The tool works by taking a snapshot of the page's accessibility tree, which assigns a short reference (e.g., @e1) to each element. You can then issue commands such as click or type using those refs, avoiding the need to re‑query the DOM. The snapshot format is plain text, not JSON, and typically consumes a few hundred tokens, which helps keep AI prompts short. Commands cover navigation, form filling, screenshots, network control, storage handling, tab management, and debugging. For example, `agent-browser open example.com` launches a page, `agent-browser snapshot -i` prints the tree, and `agent-browser click @e2` activates a link.
Agent Desktop is aimed at developers building autonomous agents that need reliable browser interaction. It supports long‑running sessions with persistent profiles, cookies, and proxy settings, so an agent can maintain authentication state across multiple commands. The client‑daemon architecture means the Rust CLI parses user commands while a native daemon keeps a Chrome instance alive, reducing startup overhead. The daemon shuts down after an hour of inactivity unless you adjust the idle timeout or run a headed browser, Safari, or iOS WebDriver session, which are exempt.
Installation is available via npm (`npm install -g agent-browser`), Homebrew (`brew install agent-browser`), or a direct binary download. The tool runs on macOS (ARM64 and x64), Linux (ARM64 and x64), and Windows (x64). It advertises compatibility with a range of LLM back‑ends, including Claude, OpenAI Codex, Google Gemini, and others that can execute shell commands. Documentation includes examples for navigating pages, taking screenshots, and using the ref‑based workflow.
Key features
- Compact text snapshots of the accessibility tree with deterministic refs
- Over 50 commands for navigation, form interaction, screenshots, network, and storage
- Persistent sessions with profiles, cookies, proxy, and auth state
- Built‑in video recording, streaming, and profiling tools for debugging
- Cross‑platform native binaries for macOS, Linux, and Windows
- Works with many AI agents that can run shell commands
Details
- Pricing
- Free
Suggest an edit
Sign in to suggest an edit.
Pros and cons
Pros
- Text‑only output reduces token usage for LLM prompts
- Ref‑based element selection provides deterministic interaction
- Persistent state supports long‑running automation tasks
- Native Rust implementation offers performance and cross‑platform support
Cons
- Requires Chrome and a native daemon; no mention of support for other browsers besides Safari/iOS WebDriver sessions
- Pricing information is not provided on the site
- Documentation excerpts are brief; detailed usage guides may need additional lookup.
Alternatives
ComposioComposio is a platform that gives AI agents secure access to external apps and tools.
EnconvoEnconvo is an AI agent for macOS that lives across your Mac.
AI Webcam Effects For Video MeetingsWebcam Effects is a browser extension that adds AI video effects to your webcam in any browser-based video call or stream.
Linkly AILinkly AI is a desktop app that turns the files on your computer into a private knowledge library that AI assistants can search.
AdminforthAdminForth is an open-source, self-hosted framework for building back-office and admin panels, released under the MIT license.
LakesailLakeSail is a data and AI platform built around Sail, an open-source compute engine written in Rust that is compatible with Apache Spark.
CerebriumCerebrium is a serverless infrastructure platform for real-time AI workloads.
SpeechpulseSpeech recognition and translation software for real-time typing, transcription, and subtitle generation.
Kilo AIKilo, also known as Kilo Code, is an open-source AI coding agent.
Digits Accounting ReimaginedDigits is accounting software built around AI.
MeshyMeshy turns photos, sketches or text into 3D models in under a minute, with AI texturing, auto-rigging, animation and plugins for Blender and Unity.
TavilyTavily is a search engine built for AI agents and large language model applications.