Kimi Work
Runs on the user's machine: mounts folders, drives a logged-in browser, runs on a schedule.
Verified: Sep 13next price check: Dec 12
Facts with sources
- Data stays local, but the agent acts inside your browser sessions and sees mounted folders in full. Source (in Russian)
- A mode for up to 300 subagents is available from the Allegro tier at $99. Source (in Russian)
- Moonshot sells access from Russia. Source (in Russian)
Strengths
- Native app for Apple-silicon Macs and Windows: reads and writes selected local folders with change summaries and rollback.
- WebBridge drives tabs in your already logged-in Chrome over the DevTools Protocol, so session cookies never leave the machine.
- Three approval modes (default, manual, fully automatic); scheduled and event-driven tasks, and a Goal mode that runs up to 24 hours.
- A gallery of office skills ready to use: data processing, pricing, copy editing and content writing.
Weaknesses
- Sign-in requires a phone number; the Helpfolder reviewer reads that as binding the user to the service.
- Unless you change the project folder, every run drops files on the C drive by default, and space fills up even on small tasks.
- Agent Swarm needs Moderato, Goal mode Allegretto, 1M-token chats only Allegro; credits are shared with Deep Research, PPT and Kimi Code.
- Em Dash never opened Kimi Work in six months of paying for Kimi: its built-in integrations target Chinese work software he does not use.
Example from practice
A cricket World Cup results spreadsheet
The reviewer created a project in a folder of his own choosing so files would not pile up on the C drive, gave the agent full permission, and asked for a spreadsheet of the last ten cricket World Cups with the number of matches each team won against each other. The request was sent from the Work tab; the video shows the setup and folder choice, and the finished spreadsheet is not examined on screen.
source: Helpfolder, Kimi Work – Desktop AI App for Office Automation
What reviewers say
Reviewer opinions on their own tasks and hardware, not measurements. Links go to the videos; transcripts live in docs/raw-data.
- Kimi Work – Desktop AI App for Office Automation · Helpfolder, Aug 20The reviewer sees Kimi Work as an office layer over the regular Kimi chat, much like Antigravity or Claude Cowork, and warns to change the project folder first.
- Inside the Kimi AI Ecosystem: MoE Models, PARL Swarms, & Developer APIs · Jengo, Sep 2The presenter frames Kimi Work as a "digital employee" on the local workstation with file, Python and logged-in Chrome access; an explainer with no hands-on test.
- Kimi Code Plan after 4 months of use (Honest Review) · Em Dash, Sep 4A six-month Kimi subscriber who never tried Kimi Work, and suspects these tools are far more useful inside China than outside it.
What it can do for you
- Morning briefing: open the tracker and chat, gather what changed overnight
- Weekly weekly.md summary on schedule
- A sweep of twelve vendor sites with a table as output
Models in this agent
Score is from our manager-task benchmark; models without a score are not measured yet.
How to brief an agent like this
The four-line brief is the Foundation module: prompt structure, LLM agents and MCP, an AI policy for the team. It starts with the free open module.