For creators and knowledge workers drowning in spreadsheets and PDFs, Moonshot AI has finally cracked the code on local execution. It has launched Kimi Work, a desktop agent that lives on your machine rather than in the cloud. Available now for macOS and Windows, this tool bypasses the usual latency of server-side processing. It reads your local files, controls your actual browser window, and handles scheduled tasks without needing to upload your data to a vendor’s sandbox. This shift means your sensitive documents and live work sessions stay under your control, addressing the primary bottleneck for professionals: access to their own data.
Unlike the agent tools that dominated the last two years-where you submit a goal and a remote server spins up a temporary environment-Kimi Work operates directly on your hardware. It reaches into the files and applications you use daily, removing the friction of data transfer and context switching.
What is Kimi Work?
Kimi Work is a standalone downloadable application, distinct from a standard web chat interface. You input goals in plain language, and the software executes actions directly on your operating system. Independent community reports indicate the app leverages Kimi K2.6, Moonshot’s flagship large language model.
Released on April 20, 2026, K2.6 is an open-weight Mixture-of-Experts architecture. It activates approximately 32 billion parameters per token and boasts a massive 256K-token context window, designed specifically for complex, multi-step workflows.
How Kimi Work Operates
The system is built on four core pillars. Understanding these mechanics clarifies the scope of what the tool can achieve.
- Agent Swarm: The application can deploy a parallel network of sub-agents on your hardware. Moonshot’s release notes state the swarm scales up to 300 sub-agents simultaneously. The architecture decomposes complex tasks into smaller parts before coordinating the final output. Documentation for K2.6 confirms the swarm can manage up to 4,000 coordinated steps.
- WebBridge: This browser extension allows the agent to interact with your internet browser as a human would. It performs searches, scrolls through pages, extracts data, and fills forms across multiple tabs. Crucially, because it hooks into your actual browser session, it inherits your existing logins and cookies, eliminating the need for repetitive authentication.
- Cron scheduling engine: A native scheduler handles jobs on a daily, hourly, or conditional basis. Moonshot specifies that triggers can include LLM agent calls or standard Python and shell scripts. A “Keep Computer Awake” toggle ensures overnight tasks do not stall due to sleep mode.
- Local files and code: The agent scans folders you mount to the system and executes Python scripts in the background. Per the release, original source files remain untouched unless you explicitly approve a modification.
The desktop application also includes pre-integrated financial data feeds. It connects directly to market data for A-shares, Hong Kong stocks, and US equities, removing the need for custom API configuration. Researchers can convert finished analysis directly into PowerPoint decks or Excel spreadsheets.
Use Cases With Examples
- Document triage: Direct the agent to a folder containing quarterly PDFs. Instruct it to summarise the contents into a single master document while preserving the originals. The swarm assigns one reader agent per file, then merges the findings.
- Web data collection: Command WebBridge to pull historical pricing for three specific stock tickers. The tool opens your browser, sets the date range, and extracts the data tables. A Python script then normalises the columns and exports an Excel workbook.
- Scheduled briefings: Configure a 7:00 AM job in the cron engine. Each morning, the system gathers headlines and drafts a markdown briefing. With the “Keep Computer Awake” feature enabled, the job persists through the night.
- Office generation: Request a short market-brief deck following a research pass. The agent drafts sections in parallel and renders native slides.
Kimi Work vs Cloud Agents
The fundamental distinction lies in execution location and data reachability. The comparison below contrasts Kimi Work with a standard cloud-based agent.
Execution location: Kimi Work runs on your desktop, whereas cloud agents operate on vendor servers.
File access: Kimi Work mounts your local folders; cloud agents require files to be uploaded or sandboxed.
Browser: Kimi Work uses your real, logged-in browser via WebBridge; cloud agents rely on hosted virtual browsers.
Scheduling: Kimi Work includes a built-in cron engine; cloud agents often depend on external tools or have limited scheduling.
Underlying model: Kimi Work reportedly runs on Kimi K2.6; cloud agents use the vendor’s hosted model.
Setup: Kimi Work requires installing the app and granting folder access; cloud agents offer zero-install access via a browser tab.
Security responsibility: Security falls on the user for local execution; the vendor assumes responsibility for cloud execution.
Neither approach is inherently superior. Local execution secures your data on-device and grants access to real files, while cloud execution trades that control for the convenience of zero-setup installation and managed safety protocols.
Scheduling: The Cron Engine in Practice
Kimi Work is driven by natural language rather than a public API. Its scheduler functions as a standard cron engine, accepting conventional cron syntax. The five required fields are: minute, hour, day-of-month, month, and day-of-week.
# Standard cron schedules the engine understands
0 7 * * * # every day at 07:00
0 * * * * # every hour, on the hour
30 8 * * 1-5 # 08:30 on weekdays only (Mon-Fri)
0 0 1 * * # 00:00 on the first day of each monthYou pair a schedule with a plain-language task. A daily briefing job might be defined as follows:
Schedule: 0 7 * * * (every day at 07:00)
Task: "Draft today's market briefing and save it to
~/KimiWorkspace/briefing.md. Ask before writing."The approval gate applies to this write action and to any subsequent web interaction.
Key takeaways
- An “Ask before acting” gate, with YOLO mode disabled, prompts for confirmation before any file write.
- Kimi Work is a local desktop agent available for macOS (Apple silicon) and Windows.
- An Agent Swarm can execute up to 300 sub-agents in parallel directly on your machine.
- WebBridge controls your logged-in browser, supported by a built-in cron engine for scheduled jobs.
- The tool reads local folders and runs Python, retaining originals unless you approve changes.
