Moonshot AI Launches Kimi, Local Desktop Agent Reportedly Running on Kimi K2.6 With 300-Sub-Agent Network
Moonshot AI introduced Kimi Work, an AI agent running on your desktop. The Beijing-based AI organization announced it this week along with downloads for macOS and Windows. Kimi Work reads local files, calls your native browser, and performs scheduled tasks. It targets information workers who have problems accessing files and live sessions.
Most agent tools of the past two years run in the cloud. You type in a target, the remote server scans the sandbox, and the hosted browser runs. Kimi Work works locally instead, accessing files and sessions you already use.
What is Kimi Work?
Kimi Work is a downloadable app, not a web chat. You give the terms in simple language, and it works on your device. The independent community cites a report that it uses the Kimi K2.6, the flagship model of the Moonshot.
K2.6 is an open-weighted Mixture-of-Experts model released on April 20, 2026. It unlocks up to 32 billion parameters per token. It holds a total 256K token window for a long, multi-step operation.
How Kimi works
Four building blocks define a product. Self-awareness helps you think about what you can do.
- Agent Swarm: Kimi Work can run multiple sub-agents concurrently on your machine. According to Moonshot’s release, the network is up to 300 sub-agents. The system divides the work into parts, and connects the results. The K2.6 mass is documented for up to 4,000 combined steps.
- WebBridge: This browser extension allows an agent to use the browser as a human. It searches, scrolls, extracts data, and fills forms in all tabs. Because it uses your actual session, it receives your existing logins and cookies.
- Cron scheduling engine: The built-in scheduler runs tasks daily, hourly, or by criteria. For each Month’s firing, triggers include calls to LLM agents and Python or shell scripts. Turning on “Keep Computer Awake” keeps nighttime activities from stopping.
- Location files and code: The agent reads the folders you install and runs Python in the background. According to the Moonshot release, the original files are always there unless you approve the change.
The desktop application also sends financial data. It is pre-compiled with market data for AA stocks, Hong Kong stocks, and US equities. According to the Moonshot release, this removes the need for a custom API setup. Completed research can be converted into PowerPoint decks or Excel sheets.
Use Cases with examples
- Document determination: Identify the agent in the quarterly PDFs folder. Ask him to condense them into one document, keeping the original intact. The swarm assigns one reader to each file, and aggregates the findings.
- Web data collection: Tell WebBridge to pull the historical values of three ticks. It opens your browser, sets the date range, and outputs the tables. Python then normalizes the columns and writes the Excel workbook.
- Organized summaries: Define the 7:00 AM job in the cron engine. Every morning it gathers headlines and writes down information. With “Keep Computer Awake”, work comes to life overnight.
- Office production: Request a short market deck after a survey pass. The agent writes sections in parallel and renders native slides.
Kimi Work vs Cloud Agents
The main difference is where the agent works and what it can access. The table compares Kimi Work against a traditional cloud agent.
| Size | Kimi Work (local) | A typical cloud agent |
|---|---|---|
| A place to do it | Your desktop | Vendor servers |
| File access | Mounts your local folders | Files are uploaded or sandboxed |
| Browser | Your real browser, signed in with WebBridge | Hosted virtual browser |
| Editing | Built-in cron engine | It is usually external or limited |
| Basic model | For me K2.6, reports | Dealer-owned model |
| Set up | Install the app, give the folder access | Zero-install, open tab |
| Security liability | It falls to the user | It falls to the dealer |
No method is completely successful. Localization saves data to your device and accesses the original files. Cloud-based transaction management makes it easy to set up and manage security.
Scheduling: The Cron Engine in Practice
Kimi Work is powered by natural language, not a public API. Its scheduler is a cron engine, so it accepts standard cron schedules. The five 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 the schedule with the task in plain language. The daily notification task reads 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 authorization gateway then applies to that script, and to any web action.
Key Takeaways
- “Ask before acting” gate, YOLO mode is closed, before any file is written.
- Kimi Work is a local desktop agent for macOS (Apple silicon) and Windows.
- Agent Swarm runs up to 300 sub-agents concurrently on your machine.
- WebBridge calls your logged in browser; The built-in cron engine runs scheduled jobs.
- It reads local folders and uses Python, keeping the original without you having to approve changes.
Marktechpost’s Interactive Explainer
Check it out Details here. Also, feel free to follow us Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to Our newspaper. Wait! are you on telegram? now you can join us on telegram too.
Need to work with us on developing your GitHub Repo OR Hug Face Page OR Product Release OR Webinar etc.? contact us
