Taskhomie
A local AI desktop assistant that operates your screen, browser, and terminal from natural-language instructions.
Per-dimension scores and reasoning
Evidence shows: The repository provides no documentation or implementation for permission management, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. All trust-related criteria are unsupported, hence score 0.
Evidence shows: README and package.json are mostly consistent in project name and description, but there is inconsistency (README title is Taskhomie, while repo name is computer-agent). Dependency list exists, but no lock file or integrity checks are provided. No error handling or failure message documentation. Thus, self-consistency scores 1, dependency availability scores 1, failure messages scores 0.
Evidence shows: README describes two modes (computer use and background) and provides use cases. Capability boundaries are implied in mode descriptions but not explicit. Triggers are defined via shortcuts but not detailed. Environment requirements (Rust, Node, macOS permissions) are listed. Thus, audience and scenarios score 1, capability boundaries score 1, trigger precision score 1, environment fit score 1.
Evidence shows: README provides installation instructions, shortcuts, and mode descriptions, but lacks FAQ, known limitations, changelog, and version history. License is Apache 2.0 with additional terms. Maintenance responsibility is implied by author but not explicit. Thus, information architecture scores 1, install notes scores 1, naming stability scores 1, examples and FAQ scores 1, known limitations scores 1, license scores 1, versioning and changelog scores 0, maintenance responsibility scores 1.
Evidence shows: README describes output (controlling computer) but no details on output format or usability. Marginal value is implied via demo and mode descriptions but not quantified. Cost-benefit is not discussed. Thus, output usability scores 1, marginal value scores 1, cost-benefit scores 1.
Evidence shows: Claims in README (e.g., controlling computer) are not traceable to evidence. No external sources or tests to corroborate. Facts and inferences are not clearly separated. Thus, claim traceability scores 1, cross-source corroboration scores 0, fact-inference separation scores 1.
- The repository provides no documentation of security mechanisms such as permission control, user confirmation, or data flow transparency; assess risks before use.
- The license includes additional terms that may affect derivative works; read carefully.
- Project name inconsistency (README says Taskhomie, repo says computer-agent) may cause confusion.
- No dependency lock file is provided, so supply chain security cannot be assured.
What does this agent do, and when should you use it?
Taskhomie is a local desktop AI agent built with Tauri, React, and Rust. It accepts natural-language instructions and can take screenshots, move the mouse, click, type, and run terminal commands. Computer Use Mode takes control of the screen, cursor, and keyboard for tasks spanning applications, while Background Mode automates the web through Chrome DevTools Protocol and uses the terminal for other work without taking over input devices. It runs locally as a Tauri app in development or production builds, but model access requires an Anthropic API key; the UI offers Haiku, Sonnet, and Opus selections.
After a user provides a natural-language task, the app started with npm run tauri dev can run one of two workflows. Computer Use Mode reads the current desktop through screenshots, then moves the mouse, clicks, types, and can run terminal commands; it directly controls the cursor and keyboard. Background Mode uses Chrome DevTools Protocol for web automation and the terminal for everything else, so it does not touch the mouse or keyboard. Shortcuts are ⌃⇧C for Computer Use Mode, ⌃⇧B for Background Mode, ⌘⇧H for help mode, and ⌘⇧S to stop the agent.
- An individual completing a task across several desktop applications can use Computer Use Mode to let the agent act from what it sees in screenshots.
- A developer working across a browser and terminal can run Background Mode for web automation and terminal work while retaining use of the mouse and keyboard.
- A macOS user who wants to start a spoken task quickly can use
⌃⇧Cor⌃⇧Bto choose foreground or background execution. - A user who wants a screenshot-based quick prompt can invoke help mode with
⌘⇧H. - A user supervising desktop automation can stop the agent immediately with
⌘⇧S.
What are this agent's strengths and limitations?
- It offers two distinct execution models: direct desktop control in Computer Use Mode and input-device-free Background Mode.
- Background Mode explicitly combines Chrome DevTools Protocol web automation with terminal-based work.
- The Tauri, React, Rust, and Tokio stack supports both
npm run tauri devdevelopment runs andnpm run tauri buildproduction builds.
- An Anthropic API key is required, creating a dependency on Anthropic as the model provider.
- Computer Use Mode directly controls the screen, cursor, and keyboard, so those devices must be relinquished while it runs.
- macOS requires Accessibility permission, and the supplied material does not document support for other operating systems.
- The license information conflicts: repository metadata says NOASSERTION while the README links to Apache License 2.0.
How do you install or deploy this agent?
Requirements: Rust and Cargo, Node.js and npm, and an Anthropic API key.
Run:
npm install
echo "ANTHROPIC_API_KEY=your-key-here" > .envFor development: npm run tauri dev
For a production build: npm run tauri build
On macOS, grant Accessibility permission when prompted in System Settings → Privacy & Security → Accessibility.
How do you use this agent?
Start the app and give it a natural-language task. Press ⌃⇧C to use push-to-talk with Computer Use Mode, which directly operates the screen, mouse, and keyboard. Press ⌃⇧B for Background Mode, which uses Chrome DevTools Protocol and the terminal without controlling input devices. Use ⌘⇧H for a screenshot plus quick prompt, and ⌘⇧S to stop the agent.
How does this agent compare with similar options?
The project presents two internal alternatives: Computer Use Mode directly controls the desktop, while Background Mode runs asynchronously through Chrome DevTools Protocol and the terminal without using the mouse or keyboard.
FAQ
Which provider and credentials does it require?
Will it take over my mouse and keyboard?
Does macOS need extra permission?
How do I stop a running task?
⌘⇧S to stop the agent.