Agent.exe is live · free · last verified 21 Oct 2026 · viability monitor
Agent.exe
Open-source Electron app that lets Claude's computer-use API drive your Mac's mouse, keyboard, and screen.
Agent.exe is 27 commits of well-organized reference code, not a product, and that's the whole reason to open it. The 3.5k-star, 320-fork signal tells you what people are doing with it: reading it, forking it, and building their own thing. If you want a maintained Mac automation agent with sandboxing and permission scopes, this isn't it — the author says so himself and only commits to reviewing PRs. If you want to understand how Claude's computer-use API drives a cursor and takes screenshots in under an hour, clone it today.
Verdict generatedVerified 8d ago · liveness 54/100 · cite: rightaichoice.com/tools/agent-exe
- Developers who want to read a minimal, working implementation of a computer-use agent
- AI researchers prototyping autonomous desktop workflows from a forkable base
- Engineers evaluating Claude's computer-use API before committing to a heavier framework
- Hobbyists with Node.js and an Anthropic API key who want an agent driving their Mac today
- Anyone needing a maintained, supported product — the author states he won't maintain it
- Teams requiring sandboxing, permission scopes, or safety guardrails around agent actions
- Non-technical users without Node.js and API key setup experience
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Agent.exe if you need a maintained product with sandboxing, permission scopes, or vendor support — the author wrote it in six hours and states he does not intend to maintain it or merge pull requests.
You pay Anthropic's token rates directly for every screenshot and action the agent takes, and a multi-step task like the README's flight booking burns tokens per step with no cap.
Agent.exe is MIT-licensed and costs nothing to clone; your only spend is Anthropic API tokens, billed at Anthropic's rates rather than a flat subscription. That puts it below any paid desktop-automation subscription, but you take on maintenance yourself — this is a fork base, not a service, and there's no tier to upgrade into.
In short
Agent.exe — Open-source Electron app that lets Claude's computer-use API drive your Mac's mouse, keyboard, and screen. Best for Developers who want to read a minimal, working implementation of a computer-use agent, AI researchers prototyping autonomous desktop workflows from a forkable base, Engineers evaluating Claude's computer-use API before committing to a heavier framework. Free to use.
What is Agent.exe?
Agent.exe — Open-source Electron app that lets Claude's computer-use API drive your Mac's mouse, keyboard, and screen. Best for Developers who want to read a minimal, working implementation of a computer-use agent, AI researchers prototyping autonomous desktop workflows from a forkable base, Engineers evaluating Claude's computer-use API before committing to a heavier framework. Free to use.
Agent.exe is a small open-source Electron app that lets you watch Claude's computer-use API actually drive a real desktop. You clone the repo, run npm install, rename .env.example to .env, paste an Anthropic API key, and run npm start. From there you type a task into the prompt box and the model moves your cursor, types on your keyboard, and takes screenshots of your primary display to figure out what to do next. The README's own demo is a flight booking. Author Kyle Corbitt describes it as a proof of concept written in about six hours and states he does not intend to maintain it, though the repo's activity page notes he will review and merge PRs that seem cool. There is no sandboxing, it only drives the primary display, and the README recommends Firefox outright because Claude behaves noticeably better there. Windows and Linux should theoretically run since the dependencies are cross-platform, but macOS is what's tested. The project is MIT-licensed, has no paid tiers, and you pay Anthropic's token rates directly for whatever the agent does. Treat it as a readable reference implementation of how desktop computer-use agents are wired, and as a fork base if you're building one.
Key features
- Claude 3.5 Sonnet computer-use API integration
- Cursor movement and mouse click simulation
- Keyboard typing and key press automation
- Screen capture of primary display fed back to the model as visual context
- Local Electron desktop app, tested on macOS
- Prompt-driven task execution (flight-booking demo in the README)
- Stop button to halt a run mid-task
- Four-command setup: git clone, npm install, .env API key, npm start
- MIT-licensed source with 320 forks for forking and rebuilding
- No sandboxing — the model runs at your machine's access level
- Primary display only
- Claude performs noticeably better in Firefox, per the README
- Theoretically cross-platform since all deps are cross-platform (Windows/Linux untested)
- Bring your own Anthropic API key; you pay Anthropic's token rates directly
- Repo activity: 27 commits, 3.5k stars, 320 forks, 22 watchers
Behind the verdict
Agent.exe earns its 3.5k stars by being exactly what it claims to be: the shortest path from a text prompt to watching Claude drive your actual cursor. The interesting engineering is all in the loop — Claude 3.5 Sonnet receives a screenshot of your primary display, emits a cursor move, a click, a keystroke, or a typed string, the Electron app executes it against your OS, and the next screenshot closes the loop. Reading that loop in a small, MIT-licensed TypeScript codebase is worth an afternoon for anyone evaluating computer-use APIs. Strengths: the setup is four commands and the README is unusually honest about limits. The stop button is a real safety valve that the author says made a semi-auto confirmation mode unnecessary because each step runs slow enough to catch manually. Firefox support is called out with a reason, not a vague recommendation. Weaknesses are the flip side of the six-hour scope. There's no sandboxing, no permission scopes, no multi-display support, and no maintenance guarantee. The repo is 27 commits with 26 open issues and 4 open PRs, which tells you the queue is not moving. Anything destructive the model decides to do executes with your user privileges. Where it fits: engineers evaluating Claude's computer-use API, researchers prototyping autonomous desktop workflows, and anyone who wants a forkable base before committing to a heavier framework. Where it doesn't: teams that need guardrails, audit logs, or a vendor to call when it breaks.
Models under the hood
as of 2026-10-25
Limitations
- The README labels Agent.exe a proof of concept and the author states he does not intend to maintain it or merge pull requests, so expect to fork and harden it yourself.
- There is no sandboxing — the AI can take destructive actions on your machine with you as the only guardrail.
- It works only on the primary display, runs on macOS with Windows and Linux described as theoretical because the dependencies are cross-platform, and Claude performs noticeably better in Firefox than other browsers.
- Setup requires an Anthropic API key, and the repo notes you must have access to the computer-use capability.
as of 2026-10-10
Researching Agent.exe? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Agent.exe still active?
Agent.exe is live · free · last verified 21 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is Agent.exe? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What do users say about Agent.exe?
How we verified this
Every listed fact about Agent.exe— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 21 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Agent.exe 11 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed pricing page
- — changed pricing page, limitations
- — changed pricing page
- — changed pricing fit, our verdict, description, who should skip it, features, switching in, switching out, limitations, hidden costs, setup time
- — changed who it suits, description, our verdict, features, who should skip it, pricing tiers, summary
- — changed pricing page
Showing the 6 most recent of 11 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Agent.exe cost in 2026?
Agent.exe: Free to use. Pricing last verified 21 Oct 2026.
Plans compared
Every published Agent.exe tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Open Source | Free | — | Developers, researchers, and hobbyists with Node.js and an Anthropic API key who want to read or fork a working computer-use agent.Adds: Starting and only tier — MIT-licensed source, no subscription, and you pay Anthropic's token rates directly. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Agent.exe's pricing actually pencils out — and where peers do it cheaper.
Agent.exe is MIT-licensed and costs nothing to clone; your only spend is Anthropic API tokens, billed at Anthropic's rates rather than a flat subscription. That puts it below any paid desktop-automation subscription, but you take on maintenance yourself — this is a fork base, not a service, and there's no tier to upgrade into.
Has Agent.exe's pricing changed?
No plan price for Agent.exe has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Agent.exe?
Pick Agent.exe if you are Developers who want to read a minimal, working implementation of a computer-use agent, AI researchers prototyping autonomous desktop workflows from a forkable base, Engineers evaluating Claude's computer-use API before committing to a heavier framework, Hobbyists with Node.js and an Anthropic API key who want an agent driving their Mac today.
Look elsewhere if you are Anyone needing a maintained, supported product — the author states he won't maintain it, Teams requiring sandboxing, permission scopes, or safety guardrails around agent actions, Non-technical users without Node.js and API key setup experience, Multi-display or multi-user environments; primary display only, single local machine.
Skip it if Skip Agent.exe if you need a maintained product with sandboxing, permission scopes, or vendor support — the author wrote it in six hours and states he does not intend to maintain it or merge pull requests.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Agent.exe actually fits — and what changes day-one when you adopt it.
You clone the repo, run npm install, paste an Anthropic API key into .env, run npm start, and type 'book me a flight to Austin' to watch Claude 3.5 Sonnet move your cursor and take screenshots of your primary display.
Outcome: You see the full screenshot-action loop in a small TypeScript codebase within an hour and can decide whether to build on the computer-use API or pick a heavier framework.
You fork the repo and start wiring in confirmation steps or logging around the cursor and keyboard calls, using the stop button to kill runs that wander off.
Outcome: You have a working skeleton with a real screenshot-action loop instead of starting from an empty project.
You point the prompt box at a repetitive browser task in Firefox — the README recommends Firefox because Claude behaves noticeably better there — and let it grind through the steps on your own machine.
Outcome: The task runs locally with your API key, no hosted service in the loop, and you can stop the run the moment it gets confused.
Use cases
- Watch an AI book a flight end-to-end from a single text prompt, as shown in the README demo video.
- Test how well Claude performs multi-step desktop workflows across different applications.
- Prototype an AI assistant that manages files, sends email, and browses the web locally.
- Run AI-driven automation scripts entirely on your own machine instead of a hosted service.
- Study or fork a readable reference implementation of the computer-use API before building your own agent.
Setup time & first value
How long it actually takes to get something useful out of Agent.exe — broken out by persona, not the marketing-page minute.
For a developer with Node.js installed: about 15 minutes to clone the repo, run npm install, rename .env.example to .env, add an Anthropic API key, and run npm start. First successful task run depends on Anthropic API access and whether you have Firefox installed, since the README says Claude behaves noticeably better there.
Switching to or from Agent.exe
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a hosted desktop-automation service: fork Agent.exe and re-point your prompts at the local prompt box, accepting that runs now use your own Anthropic API key and your machine's permissions.
- ↗To a maintained open agent framework: port your fork's changes onto a project that accepts PRs, since the author of Agent.exe has stated he does not intend to maintain it.
- ↗To Claude's own computer-use tooling: rebuild your prompt flows against the vendor's reference project if you need something heavier than this six-hour proof of concept.
Resources & guides
Tutorials & Learning
NEW Agent.exe Claude Computer Use AI Agent UI! 🤖 Open Source - Controls Computer Claude Sonnet 3.5
Josh Pocock
NEW Claude AI Update + Agent Exe is TERRIFYING...
Julian Goldie SEO
Run Locally Agent.exe Anthropic's Claude Sonnet 3.5 Computer Use Control Tutorial
AI Motion Studio
YouTube returned 6 videos for “Agent.exe”, and we withheld 3: 3 did not mention Agent.exe. Showing the 3 we can prove are about Agent.exe.
Official links
What are the best Agent.exe alternatives?
The closest alternatives to Agent.exe we list are Clevrr Computer, Desktop and Mobilerun. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Clevrr Computer | Free | Free | — |
| Desktop | Free | Free | — |
| Mobilerun | Freemium | Pricing not published | — |
Alternatives to Agent.exe
View all
Clevrr Computer
Clevrr Computer is a free, MIT-licensed open-source AI desktop agent that drives mouse, keyboard, and screenshots via PyAutoGUI.
Tools that pair well with Agent.exe
Common stack mates teams adopt alongside Agent.exe, with the specific reason each pairing earns its keep.
Clevrr Computer
Clevrr Computer is a free, MIT-licensed open-source AI desktop agent that drives mouse, keyboard, and screenshots via PyAutoGUI.
Desktop
Open-source MCP server that lets AI coding agents drive your own logged-in browser sessions locally.
Mobilerun
Mobilerun is an open-source Python framework for AI mobile automation that drives real iOS and Android phones from plain-English goals.
Agent.exe vs its closest rival
Agent.exe's closest head-to-head is with Locus Robotics — If you run a high-volume warehouse and need proven, scalable AMRs with ROI in months, Locus Robotics is the clear choice. If you're a developer exploring AI desktop automation on a budget, Agent.exe offers free, local experimentation — but it's unmaintained and unsafe for production. Pick based on your need: physical logistics vs. digital autonomy. Verdict generated
Head-to-head comparisons
Agent Exe vs Locus Robotics
If you run a high-volume warehouse and need proven, scalable AMRs with ROI in months, Locus Robotics is the clear choice. If you're a developer exploring AI desktop automation on a budget, Agent.exe offers free, local experimentation — but it's unmaintained and unsafe for production. Pick based on your need: physical logistics vs. digital autonomy.
GeneratedAgent Exe vs Truleo
If you're a law enforcement agency needing to connect siloed data and automate lead generation, Truleo is the only real option despite its cost — it's purpose-built and production-ready. For developers experimenting with AI desktop automation, Agent.exe is a free, open-source playground but lacks support, safety, and polish. These tools serve completely different domains; the choice is dictated by your role, not features.
GeneratedAgent Exe vs Presto Voice
Presto Voice and Agent.exe serve completely different purposes. Presto Voice is a mature, enterprise voice AI solution for QSR drive-thrus with proven ROI, upselling, and high non-intervention rates. Agent.exe is a free, open-source experiment for developers to let Claude control their desktop. The only commonality is AI. Choose Presto if you run a QSR chain; choose Agent.exe if you're a developer exploring AI automation.
GeneratedFrequently asked questions
Categories
Topics
Used Agent.exe? Help shape our editorial sentiment research.