Kimi Work
Moonshot AI's local desktop AI agent for knowledge work — it reads your files, drives your browser, schedules jobs, and runs a swarm of up to ~300 sub-agents on macOS and Windows.
Last verified · 2026-10-23 · by Moe Ameen
Published 2026-07-20 · Updated
Short answer
Kimi Work is Moonshot AI's desktop agent for knowledge work on macOS (Apple Silicon) and Windows. It reads local files, drives your real browser through WebBridge, runs scheduled jobs with a built-in Cron engine, and can split a task across up to 300 sub-agents. It asks before modifying files or running code. Decrypt reports paid plans from $19 a month. It builds reports, decks, and spreadsheets, not social posts.
Key takeaways
- Kimi Work runs on your own machine and works with local files and logged-in browser sessions.
- Agent Swarm splits a task across up to 300 sub-agents, per Decrypt.
- A Cron scheduler runs jobs daily, hourly, or on a condition.
- Free download; paid plans start at $19 a month, and the full swarm needs $99-plus plans.
- It outputs research, decks, and spreadsheets, not video or published posts.
What is Kimi Work?
Kimi Work is a desktop AI agent from Moonshot AI, the Beijing lab behind the Kimi model series. Moonshot offers it as a downloadable app for macOS (Apple Silicon) and Windows, described as an "AI desktop application designed for knowledge workers": an agent that lives on your computer, reads your local files with your permission, drives your browser, and runs jobs on a schedule.
It is a different product from the Kimi chat assistant — where Kimi Web answers questions, Kimi Work is built to carry out multi-step tasks on your own machine. Decrypt reports that it runs on Moonshot's Kimi K2.6 model, and the product page references K2.6.
Its headline capabilities are an Agent Swarm and a set of local tools. The Agent Swarm decomposes a task and runs specialized sub-agents in parallel — Decrypt reports up to 300 sub-agents working in parallel, each handling a slice of the task. A browser-automation layer called WebBridge drives your actual local browser (searching, scrolling, extracting data, and filling forms across tabs), so it can work inside logged-in sessions that a cloud sandbox can't reach. A built-in scheduler (a Cron engine) fires jobs daily, hourly, or on a condition, running Python or shell scripts and agent calls around the clock. It can turn research into PowerPoint decks and Excel spreadsheets, and it ships pre-integrated market data for A-shares, Hong Kong stocks, and US equities.
Because it acts on your own machine, Moonshot gates real actions behind an "ask before acting" safeguard — it requests explicit authorization before modifying files or running code. The honest framing: Kimi Work is an operations-and-research agent aimed at analysts, consultants, and knowledge workers. It gathers data, drives software, builds documents, and automates desk work. It is not a content studio — it produces internal artifacts (research, reports, decks, spreadsheets, extracted data), not captioned video, branded graphics, carousels, or scheduled social posts, and it publishes to no social platform.
What can you make with Kimi Work?
- Research briefs and reports assembled from your local files, the live web (via WebBridge), and pre-integrated market data
- PowerPoint decks and Excel spreadsheets built from the agent's findings
- Structured data extracted from websites and logged-in dashboards by the browser agent
- Scheduled, recurring jobs — daily monitoring, hourly data pulls, conditional triggers — run by the built-in Cron engine
- Multi-step task runs handled in parallel by the Agent Swarm (up to ~300 sub-agents)
- Python/shell script output and file operations run locally on your own machine
How does Kompozy turn Kimi Work research into published content?
Kimi Work is strongest at the part of a content operation that happens before you ever open a design tool: the desk work. It can watch a source on a schedule, drive your browser to pull the numbers, read your local files, and hand you a tidy brief, a market summary, or a deck. What it hands you is an internal artifact — a report, a spreadsheet, a set of extracted facts. It renders no captioned vertical video, no face-locked avatar, no brand-exact carousel, and it posts to nothing. That is exactly the seam where Kompozy takes over: drop a Kimi Work brief or report in as a source, and Kompozy generates the finished, on-brand content around it — Persona Shorts and HeyGen avatar video, Clipped Shorts, Carousels and Quote Graphics via brand-exact HyperFrames, Photo Posts, a Blog Article, and an Email Newsletter — each held to one voice by your Persona Brief and sized for every platform.
The pairing is really two automation engines meeting in the middle. Kimi Work's Cron scheduler automates the research: say it pulls this week's A-shares or US-equity moves every morning and drops a summary in a folder. Kompozy's Autopilot automates the content: point it at that summary as a source and it turns each week's findings into a captioned market-recap Short, a five-slide carousel, a LinkedIn post, and a newsletter — then schedules and publishes them across the eight social platforms plus email and blog from one queue. Kimi Work does the analyst's legwork on your desktop; Kompozy does the creator's production and distribution — and because Kompozy runs its own generation, the content half keeps producing on its own cadence regardless of what the desktop agent is doing.
- Point Kimi Work at your sources — local files, the web via WebBridge, or its market-data feeds — and let its Agent Swarm assemble a research brief, report, or deck.
- Optionally schedule that job with Kimi Work's Cron engine so a fresh brief lands on a daily or weekly cadence.
- Drop the brief or report into Kompozy as a source.
- Let Kompozy generate the finished set: a captioned Persona Short or Clipped Short, a brand-exact Carousel, Quote Graphics, native Text Posts, a Blog Article, and an Email Newsletter — all in your Persona Brief voice.
- Schedule and publish the batch across TikTok, Reels, Shorts, X, LinkedIn, and more from one queue with Autopilot.
What does this look like in Kompozy?
A real Infographic photo post generated in the Kompozy workspace. It shows a format a research summary can become; it was not made from Kimi Work output.
Who is this for, and who should skip it?
Best for:
- Analysts and consultants who automate research and data pulls on their own machine
- People tracking A-share, Hong Kong, or US market data
- Teams that want PowerPoint and Excel output from agent research
Not ideal for:
- Anyone uneasy giving an agent access to local files and browser sessions
- Creators who need finished video, graphics, or scheduled posts
Frequently asked questions
What is Kimi Work?
Kimi Work is Moonshot AI's desktop AI agent for knowledge work, available for macOS (Apple Silicon) and Windows. It runs locally, reads your files with permission, drives your browser via a tool called WebBridge, schedules recurring jobs with a built-in Cron engine, and can coordinate up to roughly 300 sub-agents (its "Agent Swarm") on a task. Decrypt reports it runs on Moonshot's Kimi K2.6 model. Confirm specs on Moonshot's site.
How is Kimi Work different from the Kimi app?
The Kimi app and web version are a chat assistant you prompt for answers. Kimi Work is a downloadable desktop agent built to carry out multi-step tasks on your own machine — reading local files, driving your browser, running scripts, and scheduling jobs — rather than just replying in a chat window.
How much does Kimi Work cost?
The app is a free download, but meaningful agent features require a paid Kimi plan. Decrypt reports paid tiers starting at $19/month (Moderato), with the full 300-agent swarm on the $99/month-and-up Allegro and Vivace plans. Confirm current pricing on Moonshot's own pages, since plans were still evolving at the authoring date.
Can Kimi Work create and publish social media content?
No. It produces internal work artifacts — research, reports, decks, spreadsheets, extracted data — not captioned video, branded graphics, or scheduled posts, and it publishes to no social platform. To turn its output into finished, on-brand content across nine platforms plus blog and email, pair it with a content engine like Kompozy.
Does Kimi Work run in the cloud?
No — that is its defining design choice. It runs as a local desktop app so it can touch your real files and logged-in browser sessions, with an "ask before acting" safeguard that requires your authorization before it modifies files or runs code.
How do we know this?
We checked every figure on this page against the sources listed below on 2026-10-23 and removed or corrected any claim we could not confirm. Features come from Moonshot's Kimi Work page; the 300-agent figure, model, and plan prices come from Decrypt's report because Kimi's pricing page did not render for us. Kompozy plan and format details come from kompozy.io and kompozy.io/pricing.
Sources
- Kimi — Kimi Work product page (read 2026-10-23)
- Decrypt — Moonshot AI's Kimi Work brings 300 AI agents to your desktop (read 2026-10-23)
- Kompozy — Home (formats and platforms) (read 2026-10-23)
Related tools
- Kimi (Moonshot AI) — Moonshot AI's consumer AI assistant — Kimi Web, the Kimi app, Kimi Work, and Kimi Code, now running on the K3 model — whose new-subscription signups were paused in mid-July 2026 after demand for K3 outran its GPUs.
- Kimi K3 — Moonshot AI's new flagship frontier model — a very large, long-context, natively multimodal model that reads images and reasons over million-token inputs, positioned as the largest open-weight model from China.
- Kimi K2.7 Code — Moonshot AI's open-weight coding model — now the first open-weight option in the GitHub Copilot model picker.
- LM Studio Bionic — LM Studio's AI agent built for open models — a local-first assistant that inspects and edits code, works over your documents, and runs on models you download, connect, or call in the cloud.
- Cowork — Anthropic's Claude-powered desktop agent that reads, edits, and creates files on your computer to finish whole tasks.
- GLM-5.3 — Z.ai's coding and agent model, released August 14, 2026. It reuses the same 743B-parameter base as GLM-5.2 and gets its gains entirely from scaled post-training, with a 1-million-token context and a jump in long-horizon coding and cybersecurity work.