Kimi Work - A general-purpose local Agent launched by Moonshot AI

What is Kimi Work

Kimi Work is a general-purpose local Agent for knowledge workers launched by Moonshot AI, released alongside Kimi's latest beta Mac and Windows clients. The tool is based on the Kimi Code kernel, supports natural language-driven operation, and can autonomously break down tasks, control the browser, call tools, process local files, and deliver work outputs such as documents, spreadsheets, and PPTs on the user's computer. Kimi Work supports up to 300 sub-Agents working in parallel, achieving a leap from Coding Agent to Working Agent.

Main features of Kimi Work

Agent cluster collaboration: Supports up to 300 sub-Agents working in parallel, autonomously breaking down complex tasks.
Kimi WebBridge browser control: Uses the browser like a human, and can scrape data from dynamically loaded pages.
Skill system: Supports installing, using, and generating custom Skills, such as website building, PPT, financial analysis, and more.
Professional data source integration: Built-in financial/business databases such as Tonghuashun, Tianyancha, iFind, and Yahoo Finance.
Local file processing: Reads, creates, and organizes computer folders, and generates long deliverables (such as 128-page reports).
Scheduled tasks: Supports setting up workflows that execute automatically on a periodic basis.
PPT and visualization generation: Calls dedicated Skills to transform report content into designed presentation decks.
Ultra-long text processing: Based on the long-context capabilities of Kimi K2.6, supports task execution lasting for several hours.

How to use Kimi Work

Download and install: Get the Kimi latest beta Mac (requires Apple chip, macOS 12+) or Windows client.
Switch to Work mode: In the client's left navigation bar, switch from Chat to the Work tab.
Create a task: Click "New Task" and describe the goal in natural language.
Add context: Use "@" to reference local files or folders, and use "/" to quickly call Skills.
Execution and monitoring: Choose "Agent" or "Agent cluster" mode, and Kimi automatically breaks down the task and executes it locally.
View deliverables: After the task is completed, view generated documents, spreadsheets, PPTs, and other results in the "Projects" section.

Core advantages of Kimi Work

Zero-barrier GUI interaction: No need to open a terminal, write commands, or configure an environment; ordinary knowledge workers can use it.
Extremely strong browser control capabilities: Through the WebBridge solution, it can scrape dynamically loaded pages such as WeChat Official Account backends and Tonghuashun, which Claude, Codex, and Atlas cannot do.
Large-scale Agent collaboration: Supports 300 sub-Agents in parallel, 13 hours of continuous coding, and 4,000+ autonomous tool calls.
Professional data ecosystem: Inherits the professional databases for finance, scientific research, law, and more from the online version of Kimi Agent.
AI-native development: The product itself was developed with deep involvement from Kimi Code, with 92% of the code generated by AI, and 50,000 lines of effective code completed in just one week.
Local secure execution: Tasks run on the local computer, and data does not leave the device.

Comparison of Kimi Work with similar competing products

Comparison dimension	Kimi Work	Claude Code
Product positioning	Forknowledge workersgeneral-purpose desktop Agent (Working Agent)	Fordevelopersterminal coding Agent (Coding Agent)
Interaction interface	GUI graphical interface, natural language-driven, zero threshold	TUI terminal interface, requires writing commands and configuring the environment
Core users	Financial analysts, researchers, product managers, operations, etc.	Software engineers, programmers
Browser control	Extremely strong: The WebBridge solution can control the browser and supports dynamically loaded pages (such as WeChat Official Account backend and Tonghuashun)	Weak: The Chrome plugin cannot read some dynamic pages (such as the WeChat Official Account backend, which directly reports an error)
Agent collaboration	Supports 300 sub-agents collaborating in parallel, autonomously decomposing complex tasks	Single-agent mode, no cluster collaboration capability
Task duration	Supports 13 hours of continuous execution, 4000+ autonomous tool calls	Usually a single session, limited stability for long-duration tasks
Professional data sources	Built-infinancial/business databases such as Tonghuashun, Tianyancha, iFind, Yahoo Finance, etc.	No built-in professional data sources; APIs need to be configured manually
Deliverable types	Documents, spreadsheets, PPTs, data dashboards, visualized web pages, academic charts	Code files, scripts, technical documentation
Skill system	Supports installing/generating/using Skills (website building, PPT, financial analysis, etc.)	No Skill system; relies on users manually writing code
Local file processing	Proactively reads, creates, and organizes folders, generating long deliverables (such as a 128-page report)	Can read and write files, but requires users to specify paths and commands
Development background	Itself deeply participated in by Kimi Code,92% of code AI-generated	Officially developed by Anthropic, traditional engineering model

Kimi Work application scenarios

Financial investment: Research Buffett's holdings over the past 10 years and generate an investment strategy Skill, connect to professional data sources to analyze whether personal holdings conform to value investing.
Research assistance: Clean local simulation datasets, draw academic charts, and assist in completing paper writing based on the data.
Market analysis: Read local product plans and prototype diagrams, combine browser research on 2400+ user comments, and generate a 128-page cross-border e-commerce market analysis report.
Office automation: Batch process local files, organize folders, generate standardized reports, and convert them into PPT with one click.
Data dashboard setup: Scrape data from platforms such as WeChat Official Account backend, Tianyancha, and iFind to directly generate visualized analysis charts.