Lapu AI: Free Desktop Agent for macOS and Windows (2026)
Lapu AI is a desktop AI agent for macOS and Windows. Free plan starts at $0; paid plans from $20/mo. It automates files, shell, and apps with permission gates.
Lapu AI is a native desktop agent for macOS and Windows that runs files, shell commands, and app workflows on your computer, not in a cloud workspace. It is semi-autonomous: it can read and act on your behalf, but pauses before writing files or running destructive commands. Free plan available, with paid plans from $20/mo. Best for professionals doing repetitive cross-app desktop work.
Lapu AI is a desktop AI agent for macOS and Windows, founded in 2025. It automates file management, shell commands, and desktop app workflows with permission-gated execution, meaning you approve sensitive actions before they run. Plans start free at $0/mo and go up to $100/mo. The agent uses built-in frontier AI models, with all data staying on your machine and no cloud storage.
Maker: Lapu AI · Autonomy: semi autonomous · Maturity: GA
About Lapu AI
Lapu AI is a desktop AI agent built in 2025 that is designed to automate the repetitive computer work filling most professional days. Unlike browser-based AI assistants that only give suggestions, Lapu AI runs natively on macOS 12+ and Windows 10/11, giving it direct access to your file system, terminal, and desktop applications. The result is an agent that can execute tasks end-to-end rather than just telling you what to do next. The agent pairs frontier AI models with a set of local tools: file read, write, edit, and grep; shell command execution; and OS-level accessibility APIs that let it click, type, and navigate inside any desktop app. Every sensitive action, such as a file write, a shell command, or a click that could change state, is gated behind an explicit per-action approval prompt. Low-risk reads are auto-approved; anything destructive waits for you. Every action is logged with timestamps, tool calls, and outcomes for a full audit trail retained for 90 days. Best-fit workflows include document processing, data extraction from PDFs or spreadsheets, file organization across hundreds of items, research synthesis, inbox triage, and recurring multi-step reports. A floating mini chat overlay lets you prompt the agent without switching away from whatever you have open. Workflows can be saved as reusable skills and scheduled to run on a timer, turning one-off tasks into repeatable operations. Lapu AI is free to start with no credit card required, giving users the full desktop application, built-in frontier AI within a monthly usage allowance, and all core tools. Paid plans raise that allowance: Premium at $20/mo, Pro at $60/mo, and Max at $100/mo. Team and enterprise tiers add SSO/SAML authentication and shared usage pools for organizations. Founded in 2025, Lapu AI entered a market that split sharply in 2026 between passive screen-readers like Microsoft Recall and active execution agents. Lapu AI's position is explicit: local execution, permission before action, and a complete audit log so users can review every step the agent took on their behalf.
Pricing
Free plan ($0/mo, no credit card): full desktop app, frontier AI within monthly allowance, all core tools. Premium $20/mo, Pro $60/mo, Max $100/mo raise the AI reasoning allowance. Team and enterprise tiers add SSO/SAML and shared usage pools.
Key Features
- Permission-Gated Execution: Every file write, shell command, and desktop input pauses for explicit user approval before running, with low-risk reads auto-approved and a 90-day audit log of every action taken.
- File and Shell Automation: Organizes, renames, and searches thousands of files by type, date, or content, and runs terminal commands with chained operations without the user needing to remember syntax.
- Desktop Automation via Accessibility APIs: Controls any macOS or Windows app using OS-level accessibility APIs to list windows, take screenshots, click, type, and navigate inside apps the agent was not originally designed for.
- Reusable Skills and Scheduling: Package any multi-step workflow into a named skill that can be rerun or scheduled to trigger automatically, turning one-off tasks into recurring operations without re-prompting.
- Local-First Privacy: All files and data stay on the user's machine with no Lapu AI cloud storage or remote workspace; only the context needed for AI reasoning is sent to the model endpoint during a task.
- Floating Mini Chat Overlay: A persistent overlay lets you describe a task from any app without switching windows, so the agent starts working without interrupting what you already have open on screen.
Strengths
- Free tier includes the full desktop application and all core tools with no credit card required, unlike most desktop agents that require a paid subscription from day one.
- Per-action approval prompts for destructive operations and a 90-day audit log give more execution transparency than cloud-based agents where actions happen in a remote sandbox you cannot inspect.
- Runs on both macOS 12+ (Apple Silicon) and Windows 10/11 (64-bit), covering the two dominant desktop platforms without feature parity gaps between them.
Weaknesses
- The underlying frontier model powering the agent is not publicly named, making it impossible to compare reasoning quality against agents with published SWE-bench or GAIA benchmark scores.
- No published benchmark scores (SWE-bench, WebArena, GAIA) as of June 2026, so objective performance comparisons with coding agents like Claude Code (80.8% SWE-bench) are not possible.
- macOS support requires version 12 Monterey or later on Apple Silicon, which excludes older Intel Macs and any machine below that OS version.
Frequently Asked Questions
What is Lapu AI and what does it do?
Lapu AI is a desktop AI agent built in 2025 that runs natively on macOS 12+ and Windows 10/11. Unlike browser-based AI tools that only give suggestions, Lapu AI takes direct action on your computer: it reads and writes files, runs shell commands, and controls desktop applications using OS-level accessibility APIs. It is semi-autonomous, meaning it can chain multiple steps together but pauses before any action that could modify or delete data, waiting for your explicit approval. The agent includes a floating chat overlay that lets you prompt it from any app without switching windows. Every action is logged with timestamps and tool call details for up to 90 days. The free tier includes the full application and all core tools with no credit card required.
How much does Lapu AI cost in 2026?
Lapu AI offers four tiers in 2026: Free ($0/mo, no credit card required), Premium ($20/mo), Pro ($60/mo), and Max ($100/mo). The Free tier includes the full desktop application, built-in frontier AI within a monthly usage allowance, local file and shell tools, desktop automation, and reusable skills. When the monthly AI allowance is used up, the agent surfaces a clear message rather than degrading silently or charging automatically. Paid tiers raise the monthly AI reasoning allowance and add team features. Enterprise plans add SSO/SAML authentication and shared usage pools for organizations, with pricing available on request. All plans include the same core desktop agent capabilities; the main difference is how much AI reasoning you can use each month.
Is Lapu AI fully autonomous?
No, Lapu AI is semi-autonomous. It can plan and execute multi-step tasks without asking you to manage each individual step, but it pauses at specific checkpoints: file writes, shell commands, and desktop input all require your explicit per-action approval before they run. Low-risk read operations, like reading a file or taking a screenshot, are auto-approved to keep the workflow moving. This design places it between a fully autonomous agent (which would execute everything without asking) and a copilot (which only suggests actions for you to take). Every action the agent takes is logged in real time, so you can review exactly what happened and in what order. The company describes this as 'permissioned execution', making it one of the more conservative autonomy models in the desktop agent category.
What AI model powers Lapu AI?
Lapu AI uses frontier AI models built into the product, but as of June 2026 the company has not publicly named the specific LLM or provider powering the agent. This means users do not need to bring their own API keys or configure a model provider, which simplifies initial setup. The model is accessed through a managed endpoint operated by Lapu AI: only the task context needed for reasoning is sent to the model, while all files and data stay on the user's machine. Because the underlying model is not disclosed, independent benchmark comparisons against agents with published SWE-bench or GAIA scores are not currently possible. Users evaluating the reasoning quality should run their own tasks on the free tier before committing to a paid plan.
What are the best alternatives to Lapu AI?
The closest alternatives depend on your specific needs. Claude Code from Anthropic is a terminal-based coding agent with 80.8% on SWE-bench Verified and is the stronger choice for software engineers who work in the command line. Microsoft Copilot on Copilot+ PCs offers passive AI assistance built into Windows, but it is a suggestion layer rather than a full execution agent. Anthropic's Claude Cowork, launched in March 2026 for macOS Pro and Max users, drives the mouse and keyboard in a similar way to Lapu AI but is tied to a higher-tier Claude subscription. Lapu AI's main advantage over all of these is its combination of cross-platform desktop execution, explicit permission gating, and a free starting tier with no credit card required.
Who is Lapu AI best for?
Lapu AI is best suited for individual professionals and small teams who do high-volume, repetitive desktop work and do not want to write code or set up cloud infrastructure to automate it. Ideal users include operations managers who run the same multi-step workflows daily, analysts who extract data from PDFs or spreadsheets and move it into other apps, and non-technical users who want an AI agent that takes real action rather than only giving advice. It is not a good fit for software engineers who need a coding agent with benchmark-verified output quality, or for teams needing a cloud-hosted agent that runs 24 hours a day without an open laptop. Lapu AI requires the host machine to be on and running for any task to execute, which limits overnight or unattended automation.
How does Lapu AI compare on benchmarks?
As of June 2026, Lapu AI has not published scores on standard agent benchmarks such as SWE-bench, WebArena, or GAIA. This makes objective performance comparisons with other agents difficult. For reference, Claude Code scores 80.8% on SWE-bench Verified, which is the leading public score for a coding agent. Lapu AI is a general desktop automation agent rather than a specialized coding agent, so SWE-bench is only a partial measure of its real-world utility anyway. The company instead highlights its permission-gating model, 90-day audit logging, and local-first privacy as its primary differentiators from agents focused purely on throughput benchmarks. Users evaluating Lapu AI should run their own workflow tests on the free tier to assess whether its reasoning quality meets their needs.
How do you get started with Lapu AI?
Getting started takes around five minutes. Go to lapu.ai, download the desktop app for your platform: macOS 12 Monterey or later on Apple Silicon, or Windows 10/11 64-bit. Create a free account with no credit card needed and install the application. After installation, launch the app and a setup walkthrough guides you through granting the accessibility permissions the agent needs to control your desktop. From the floating mini chat overlay, type your first task in plain language, for example 'organize all PDF files in my Downloads folder by date.' The agent shows its planned steps, you approve or reject each sensitive action, and it completes the task. The free tier's monthly AI allowance is sufficient for dozens of typical workflows before an upgrade becomes necessary.