Moonshot AI presents Kimi Work: the AI desktop agent for macOS and Windows

Moonshot AI, one of the most promising Chinese startups in the artificial intelligence sector, has launched Kimi Work, an AI desktop agent for macOS and Windows. This new tool, currently in internal testing phase, is designed to operate directly on the user's computer, reading local files, controlling the browser, and performing scheduled tasks. The news was announced this week, with free downloads available on the official website.

An innovative approach: local AI

Most current AI tools operate in the cloud, requiring an internet connection to process user requests. Kimi Work, on the other hand, represents a radically different approach. Installed directly on the user's computer, this AI agent has access to local files and can interact with the operating system, manipulate PDF documents, organize the desktop, extract stock data from the browser, and even compile HTML reports to send via email.

Agent Swarm and WebBridge: distinctive features

One of the most innovative features of Kimi Work is Agent Swarm, which allows up to 300 sub-agents to be launched in parallel, each tasked with a specific part of a complex task. Additionally, Kimi Work is integrated with WebBridge, an extension that allows the agent to control the user's real browser via Chrome DevTools Protocol, maintaining login sessions and local cookies.

An integrated Cron engine allows tasks to be scheduled on a daily, hourly, or conditional basis, with a "Keep Computer On" option for night jobs. Additionally, Kimi Work includes a local file layer that allows the agent to read mounted folders and execute Python scripts in the background.

Integrated market data and professional output

Kimi Work is equipped with native market data for A-share stocks, Hong Kong titles, and US stocks, ready to use without the need to configure external APIs. Completed searches can be directly converted into PowerPoint presentations or Excel sheets, making the tool particularly useful for financial sector professionals.

Kimi K2.6: the model behind Kimi Work

Kimi Work runs on Kimi K2.6, an open-weight model released by Moonshot AI on April 20. K2.6 is a mixture of experts model with about a trillion parameters, which activates about 32 billion parameters per token, with a context window of 256K tokens. This means the agent can maintain a large amount of information in memory during complex and multi-step workflows, without losing sight of the initial goal.

Privacy and security: important considerations

The term "local" in Kimi Work primarily refers to the place where actions occur, i.e., the user's computer. However, the inference of the K2.6 model can still take place via Moonshot's cloud API, even if file readings, browser clicks, and Python script execution occur locally. For completely on-device inference, the model weights are available on Hugging Face under Modified MIT license, but running a model with a trillion parameters requires powerful hardware that most home users do not possess.

Privacy with Kimi Work is more nuanced than one might think. Since WebBridge controls the user's real browser, it can access bank accounts, emails, and internal corporate tools. Researchers at the University of California Riverside have warned that AI agents often perform tasks without recognizing the associated risks, a behavior defined as "blind goal-directedness". Moonshot includes an "ask before acting" mode that requires user approval before modifying files or executing code, but this precaution does not guarantee 100% security.

The AI agent war: Kimi Work in the competitive landscape

The market for AI desktop agents has quickly become crowded. Anthropic's Claude offers full desktop use from late 2024, OpenAI introduced Codex Background Computer Use for macOS in April 2026, Google's Gemini Computer Use focuses on browser workflows, and Microsoft's Copilot Studio added computer use functionality in May 2026, aiming at business automation.

However, it seems that users prefer flexible tools that can configure AI agents using any language model on APIs. Tools like OpenClaw, Hermes, or NanoClaw come into play in this context. The difference with Kimi Work lies in the local-first design combined with a 300-agent Agent Swarm. Most competitors run everything in a cloud sandbox that cannot access real login sessions, or offer desktop control without coordinated parallel agents. Kimi Work does both. The trade-off is that when the laptop is closed, tasks stop. Moonshot's cloud product, Kimi Claw, on the other hand, can run tasks 24/7 without depending on the user's computer.

Subscription plans and availability

Kimi Work is available for free download, but significant agent functionalities require a paid plan. The Moderate plan starts at $19/month and includes access to K2.6, Deep Research, and Kimi Code. Agent Swarm with a limited number of sub-agents unlocks with the Allegretto plan ($39/month), while full Agent Swarm and high-volume professional workflows require the Allegro ($99/month) or Vivace ($199/month) plans for those needing all 300 agents.

Downloads for macOS (Apple Silicon) and Windows are active on kimi.com, with the internal testing phase meaning some features may still change before a wider release.

The impact of Kimi Work on knowledge workers' jobs

Kimi Work represents a significant breakthrough for knowledge professionals, offering an unprecedented level of automation and integration. The ability to manipulate PDF documents, organize the desktop, extract stock data from the browser, and even compile HTML reports to send via email makes it particularly useful for financial sector professionals. However, the tool's adoption requires careful evaluation of security implications and technical needs. With the right approach, this tool could become an indispensable tool for professionals and companies, opening new possibilities for digital work automation.

Considerations for companies

For organizations evaluating the adoption of Kimi Work, it is crucial to consider not only operational advantages but also cybersecurity implications. Creating clear policies on the use of AI agents and training staff on digital security best practices will be key elements for successful implementation.

The role of Kimi K2.6 in the AI ecosystem

The technological foundation of Kimi Work, the K2.6 model, represents an important advancement in the architecture of language models. Its ability to maintain an extended context and manage complex workflows without loss of information makes it particularly suitable for advanced professional applications. This could open new opportunities for the development of specialized tools in various sectors.

The technical challenges and limitations

Despite its capabilities, Kimi Work has some technical limitations. Dependence on the local computer means activities can be interrupted when the device is turned off or disconnected. Additionally, implementing such a complex agent system requires significant hardware resources, which could be an obstacle for some users.

The opportunities for developers

The open architecture of Kimi Work, with model weights available on Hugging Face, offers interesting opportunities for the developer community. The possibility of customizing and extending system functionalities could lead to the development of innovative solutions and the creation of an ecosystem of plugins and integrations. Kimi Work represents a significant step forward in the field of AI agents, offering an unprecedented level of automation and integration. However, its adoption requires careful evaluation of security implications and technical needs. With the right approach, this tool could become an indispensable tool for professionals and companies, opening new possibilities for digital work automation.

Editorial Note and Disclaimer

The guides and content published on GoYou are the result of independent research and analysis activities, for informational, educational, and in-depth purposes.

GoYou does not constitute a journalistic publication or an editorial product pursuant to Law No. 62/2001 and does not provide real-time information.

The GoYou project does not provide professional, technical, legal, or financial advice and disclaims any liability for the improper use of the information published.

In the Crypto sector, every investment involves risks: the reader is invited to always inform themselves autonomously before making any decision.