Moonshot AI’s Kimi Work unleashes 300 AI agents on your desktop, no cloud required

Running 300 AI agents on your laptop sounds like a recipe for a very expensive space heater. But Moonshot AI is betting it’s actually the future of knowledge work.

The Beijing-based AI firm launched Kimi Work on June 9-10, a desktop application powered by its new Kimi K2.6 model that can coordinate up to 300 specialized sub-agents operating in parallel on a user’s local machine. That’s a threefold increase from the 100-agent ceiling of its predecessor, Kimi K2.5.

What Kimi Work actually does

Each sub-agent handles a specific slice of a larger workflow: research, document creation, coding, data analysis, browser automation.

The K2.6 model underneath is a 1-trillion-parameter Mixture-of-Experts architecture. Instead of running one massive neural network for every task, the system activates only the relevant “expert” sub-networks for each job. This is what makes running hundreds of agents on local hardware feasible rather than purely theoretical.

Kimi K2.6 can manage workflows spanning roughly 4,000 steps, up from 1,500 steps with K2.5. That’s the difference between an AI that can draft a memo and one that can research a topic across multiple sources, cross-reference data, build a spreadsheet, write a summary, and schedule a follow-up meeting.

Browser tasks get handled through a WebBridge extension, letting agents interact with web pages, fill out forms, and pull data without the user needing to manually intervene. The application runs on macOS (Apple Silicon) and Windows.

Kimi Work is designed to run locally. Your files, your browser history, your schedule, none of it routes through external servers by default.

Moonshot AI’s meteoric rise

At the end of 2025, the firm was valued at $4.3B. By May 2026, that number had ballooned to $20B. Now the company is reportedly seeking up to $2B in fresh funding that could push its valuation to $30B.

The K2.6 model has drawn attention for strong performance in agentic coding and long-horizon tasks. Moonshot AI has released open weights under a modified MIT-style license.

Moonshot AI’s internal terminology for the underlying technology is “Agent Swarm,” which describes the system’s ability to deploy and coordinate large numbers of specialized agents toward a unified goal. The practical applications include financial analysis, report generation, and project management.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.