Moonshot AI shipped Kimi Work as a downloadable desktop agent for knowledge work — not another chat tab, but a local client that reads your folders, drives your real browser, schedules Cron jobs, and coordinates K3 Agent Swarm runs into PowerPoint and Excel. The product page at kimi.com/products/kimi-work tags it "The AI Desktop for Knowledge Work" and positions it against quick web Kimi chat: a system-level digital employee for analysts, researchers, and ops teams who live in spreadsheets and logged-in web apps.
This lands the same week Kimi K3 dominated eval chatter, new K3 subscriptions paused on GPU capacity, and Western labs pushed ChatGPT desktop and Claude Cowork local-agent narratives. explainx.ai's read: Kimi Work is Moonshot's bet that swarm + real Chrome + office output + market data beats a generic coding terminal for non-engineer knowledge workers — with honest limits on Apple Silicon-only Macs, sleep-killed Cron jobs, and paid tiers for full swarm.
TL;DR — question-first reference
| Question | Answer (July 21, 2026) |
|---|---|
| What is Kimi Work? | Moonshot desktop local agent — files, browser, Python, Cron, swarm |
| Free download? | Yes — app free; paid membership unlocks agent credits + swarm |
| Platforms? | macOS (Apple silicon only) · Windows — no Intel Mac |
| vs web Kimi? | Web = chat; Work = mounted folders + WebBridge + schedules |
| WebBridge? | Chrome/Edge extension + CDP — logged-in sessions stay local |
| Swarm scale? | Up to ~300 sub-agents, 4000+ tool calls, K3 Swarm branding |
| Runs while asleep? | No — use Keep Computer Awake or Kimi Claw (cloud) |
| Finance data? | Native A-shares, HK, US equities — earnings, anomalies, reconciliation |
| Cheapest paid tier? | Moderato $19/mo — basic agents, no full swarm |
| Full swarm tier? | Allegro $99 / Vivace $199 — highest quotas per help center |
| K3 sub pause? | July 19 Moonshot paused new K3 subscriptions — capacity story |
| vs OpenClaw? | OpenClaw = open gateway + BYO models; Kimi Work = closed Moonshot stack |
| vs jcode swarm? | jcode = Rust coding harness on repos; Kimi Work = office + browser + finance |
What is Kimi Work?
Kimi Work is the desktop mode inside Moonshot's Kimi Desktop App — the same installer path that serves Kimi Work vs lighter chat surfaces. Moonshot markets three pillars on the product page:
- Local file mounting — agent reads/writes under folders you authorize (example workspace:
~/Documents/KimiWorkspace). - Browser automation — WebBridge drives real Chrome or Edge through a local extension.
- Always-on workflows — Cron engine for scheduled LLM agent calls and Python/Shell scripts.
Official copy: "Deeply connected to your local files. Capable of browser automation. Running around the clock." The emphasis is knowledge work productivity, not IDE pair-programming — though Kimi Code remains the terminal path for developers who want /model K3 selection.
Internal testing caveat: Decrypt's July 2026 launch coverage notes Kimi Work shipped alongside free downloads but was still characterized as internal testing at announcement — treat feature stability and regional availability as verify-on-install, not production-SLA gospel.
Example prompt — research to Excel
Moonshot's product page walks through a representative workflow:
Search for the latest AI trend reports, extract key insights from the first 3 pages, and organize them into an Excel spreadsheet saved to
~/Documents/KimiWorkspace.
That single sentence exercises WebBridge search, multi-page extraction, structured summarization, and local Excel creation — the pattern finance and strategy teams repeat daily without opening an IDE.
Kimi Work vs web Kimi — when to use which
Moonshot's FAQ draws a clean line:
| Surface | Best for | What it cannot do |
|---|---|---|
| Web Kimi (kimi.com) | Quick chat, ad-hoc Q&A, mobile-friendly queries | Mount ~/Documents, run midnight Python, drive logged-in broker portals |
| Kimi Work (desktop) | Deep workflows, scheduled briefings, browser scraping, swarm doc builds | Replace a full IDE for git-native coding — use Kimi Code instead |
| Kimi app (iOS/Android) | Capture context on the go, trigger consumer swarms | Local file Cron, WebBridge on phone — see mobile K3 guide |
Hybrid pattern: Mobile Kimi app for intake → desktop Kimi Work for execution mirrors how teams already split closed cloud vs local stacks. Web Kimi answers "What does this mean?" Kimi Work answers "Do this every night and put the result in my folder."
WebBridge — deep dive on real-browser automation
WebBridge is Kimi Work's differentiator vs ChatGPT desktop tabs or Cowork's sandboxed browsing. Architecture per kimi.com/features/webbridge:
Kimi Work agent → local WebBridge service → Chrome DevTools Protocol → your Chrome/Edge
↑ browser extension pairs session
Results (DOM, screenshots, extracted tables) → back to agent locally
What WebBridge can do
| Capability | Example task |
|---|---|
| Multi-tab navigation | Open three competitor pricing pages, diff features |
| Click / scroll / fill | Submit filter forms on a research database |
| Cross-site search | Google → open top results → merge notes |
| Sheet building | Scrape tables into structured rows |
| Workflow → Skill | Convert a repeated WebBridge run into a reusable skill |
Moonshot demo videos on the WebBridge page highlight Google Sheets construction, cross-site search, and workflow-to-skill conversion — all through the same logged-in browser you use manually.
Privacy and session model
Because automation hits your existing Chrome/Edge profile:
- Bank, broker, and SaaS logins stay in local cookies — the agent does not need you to paste passwords into a cloud VM.
- Page content flows through the local CDP bridge, not a Moonshot-hosted headless farm (per official FAQ copy: "Everything runs locally").
- Trade-off: you inherit real browser risk — malicious pages, prompt injection in DOM text, and accidental clicks on production admin panels. Same class of issues as Claude Cowork safety guidance — external content can steer agents.
Setup checklist
- Install Kimi Desktop App (macOS Apple silicon or Windows).
- Install WebBridge extension from the browser store (Chrome/Edge).
- Switch to Kimi Work mode in the desktop app.
- If disconnected: reinstall extension → resend connection command → restart Kimi Desktop App (official troubleshooting path).
Not Playwright-in-the-cloud: WebBridge is closer to local computer-use than remote browser APIs — philosophically aligned with OpenClaw's local-first gateway, but closed-source and Moonshot-hosted models.
Agent Swarm and Kimi K3 — parallel sub-agents to office docs
Agent Swarm is Kimi Work's answer to multi-hour research and deck-building. Moonshot coordinates specialized sub-agents in parallel, then converts output to PowerPoint or Excel without a separate export step.
K3 Swarm — current generation
The product site now brands swarm as K3 Swarm — powered by Kimi K3 (2.8T MoE, 1M context) rather than launch-era K2.6 references in early press. Help-center and marketing claims include:
| Metric | Claim |
|---|---|
| Max sub-agents | Up to ~300 parallel specialists |
| Tool calls per run | 4000+ in heavy swarms |
| Speed | ~4.5× vs single-agent mode (Moonshot marketing) |
| Output | PowerPoint decks, Excel workbooks from swarm research |
Capacity reality check: Official membership pricing gates swarm by monthly uses and concurrent subtasks — not unlimited 300-agent runs on every plan:
| Plan | Agent Swarm uses/mo | Concurrent subtasks |
|---|---|---|
| Moderato ($19) | 25 | 2 |
| Allegretto ($39) | 50 | 4 |
| Allegro ($99) | 120 | 4 |
| Vivace ($199) | 240 | 8 |
Decrypt summarized the go-to-market split: limited swarm at Allegretto, full professional swarm volume at Allegro/Vivace. The 300-agent headline is a ceiling, not your default Tuesday afternoon job.
Swarm vs coding harness swarms
For developers comparing orchestration models:
| Product | Swarm focus | Output surface |
|---|---|---|
| Kimi Work K3 Swarm | Research, web, finance, slides/sheets | PPTX, XLSX, briefings |
| jcode swarm | Same-repo coding agents, collision signals | Git commits, Rust/TS code |
| Kimi app macOS 27 demo | Consumer long-horizon build | Web app in browser |
Kimi Work optimizes knowledge deliverables; jcode optimizes repository concurrency with RAM efficiency. Different buyers, overlapping "many agents" marketing.
Cron and scheduling — Set It & Forget It
Kimi Work ships a built-in Cron engine Moonshot brands "Set It & Forget It":
| Trigger type | Example |
|---|---|
| LLM Agent Call | 6:00 AM daily market briefing draft |
| Python script | Midnight dataset normalization |
| Shell job | Hourly log aggregation from local exports |
| Conditional schedules | Run when file appears in watched folder |
Official scenarios: "early-morning LLM Agent call to draft daily briefings" and "midnight Python script to process massive datasets."
Sleep, lid-closed, and honest limits
Moonshot's FAQ is explicit: scheduled tasks need an awake machine. Toggle Keep Computer Awake in settings for overnight runs. Decrypt's coverage repeats the limitation — laptop sleep stops Cron — which matters for travelers who expect cloud cron reliability.
Cloud alternative: Kimi Claw (Allegretto tier and above per pricing table) is Moonshot's 24/7 cloud agent when your laptop is closed — the parallel to "always-on" without keeping MacBook fans spinning at 2 AM.
Built for finance — A-shares, HK, US equities
Kimi Work ships native market data integrations Moonshot highlights for finance analysts:
| Market | Use cases (per product page) |
|---|---|
| A-shares (China) | Earnings reports, anomaly scans |
| Hong Kong stocks | Cross-list comparisons |
| US equities | Earnings, reconciliation workflows |
Moonshot copy: "Skip the complex API setups — instantly pull earnings reports, analyze market anomalies, and reconcile spreadsheets through natural conversation."
Professional database calls scale with membership — official pricing shows 200 calls (free) up to 24,000/month (Vivace). That quota is separate from generic agent credits but shares the paid-tier upsell story: heavy terminal users need Allegro+, not Adagio.
Regional nuance: Moonshot is a Beijing-based AI Tiger — A-share depth is a home-market advantage Western desktop agents rarely match. US and HK coverage makes Kimi Work plausible for global macro desks already bilingual in CN/US filings — verify data freshness and licensing for compliance before production research.
Privacy model — local files and Ask before acting
Moonshot addresses the obvious enterprise objection on the product FAQ:
"You have absolute control over your files. The built-in Ask before acting safeguard means Kimi will prompt you for explicit authorization before it modifies, overwrites, or runs code within your local directories. Nothing happens without your consent."
| Layer | Behavior |
|---|---|
| File writes | Prompt before modify/overwrite/code execution in mounted dirs |
| WebBridge | Local CDP — sessions stay in your browser |
| Cloud inference | Agent reasoning still hits Moonshot servers for K3 — not fully offline |
| Enterprise path | Kimi Enterprise — separate org accounts; verify training opt-out |
This is not run K3 locally — weights promised July 27, 2026 target self-hosters, not Kimi Work's default cloud brain. Privacy here means data locality for files and browser state, not air-gapped LLM inference.
Teams blocked on Moonshot provenance debates or cyber guardrails geopolitics should treat Kimi Work like any other Chinese frontier cloud agent with local file reach — high utility, policy review required.
Pricing tiers — free app, paid agent credits
Kimi Work follows Moonshot's musical tempo membership names. Official pricing table (July 2026):
| Plan | Monthly | Annual (per mo) | Agent credits* | Swarm | Kimi Claw |
|---|---|---|---|---|---|
| Adagio (Free) | $0 | $0 | 6 | — | — |
| Moderato | $19 | $15 | 60 | 25 uses / 2 subtasks | — |
| Allegretto | $39 | $31 | 150 | 50 / 4 | ✅ |
| Allegro | $99 | $79 | 360 | 120 / 4 | ✅ |
| Vivace | $199 | $159 | 720 | 240 / 8 | ✅ |
*Approximate task equivalents — actual burn depends on K3 token usage per job.
Also included on paid tiers: Kimi Code credits (1×–30×), 4× agent speed priority, professional database call pools.
July 19 capacity pause: Moonshot paused new K3 subscriptions while splitting Kimi Membership (Web/App/Work) from Kimi Code Membership. Existing subscribers kept access; new signups reopen in batches — check status before budgeting team seats.
API alternative: kimi-k3 on platform.kimi.ai bills per token ($3/$15 per M input/output) — different economics from desktop membership credits. Hybrid teams may use Work for WebBridge + Cron and API for CI.
Kimi Work vs competitors — desktop agent landscape
| Dimension | Kimi Work | ChatGPT desktop | Claude Cowork | OpenClaw | jcode |
|---|---|---|---|---|---|
| Primary user | Analysts, researchers, finance | General knowledge worker | Document/folder agent | Hackers, messaging power users | Developers |
| Browser model | WebBridge — real Chrome/Edge | Integrated browser | Sandboxed browsing | BYO / extensions | Firefox Agent Bridge |
| Swarm | K3 Swarm → PPT/XLS | Limited multi-step | Subagent patterns | Community skills | Same-repo Rust swarm |
| Scheduling | Built-in Cron | Varies by product | Limited | Cron via host | Server mode |
| Market data | A/HK/US native | General web | General web | None built-in | None |
| Open source | No | No | No | Yes (Foundation) | MIT |
| Model lock-in | Moonshot K3/K2.x | OpenAI | Anthropic | Any provider | 40+ providers |
| Mac support | Apple silicon only | Broader | Varies | macOS/Linux/Win | macOS/Linux/Win |
explainx.ai pick logic:
- Choose Kimi Work when WebBridge + finance APIs + swarm-to-Excel beats assembling MCP tools yourself.
- Choose OpenClaw when Telegram/WhatsApp gateway + model choice + nonprofit stewardship matter more than Moonshot's integrated office pipeline.
- Choose jcode when the deliverable is merged code on GitHub, not a board deck by 9 AM.
- Choose ChatGPT Work / Codex or Cowork when Western vendor procurement, quota politics, or Anthropic safety docs dominate policy — see Fable limit extensions through August 19 for the competitive counter-move.
What people are asking — setup, region, and maturity
Does Kimi Work run on Intel Macs?
No. The download page specifies macOS (Apple silicon) only. Intel Mac users need Windows (Boot Camp / separate PC) or web/API paths — not a Rosetta workaround Moonshot advertises.
Is Kimi Work available in the US?
Moonshot operates China-first with growing international API access. Consumer kimi.com and app availability vary by region; platform.kimi.ai API keys remain the most portable US path for K3. Kimi Work desktop signup may lag Mainland China rollout — verify download and account creation in your jurisdiction before team pilots.
Can I use Kimi Work while K3 subscriptions are paused?
Existing members: protected per July 19 announcement. New subscribers: wait for batch reopenings or use API billing separately. Desktop agent features consume membership agent credits, not API keys by default.
WebBridge vs asking the model to "browse"
Web Kimi can summarize URLs you paste. WebBridge executes multi-step autonomous navigation — pagination, form filters, cart flows — inside your authenticated session. That is the difference between reading a link and operating a terminal.
Kimi Work vs Kimi Claw — which is 24/7?
| Product | Runs when laptop closed? |
|---|---|
| Kimi Work | No (unless machine awake + Keep Awake) |
| Kimi Claw | Yes — cloud agent (Allegretto+) |
Pair Work for local file + WebBridge sensitivity with Claw for overnight monitors if budget allows Allegretto or higher.
Who should use Kimi Work?
Strong fit:
- Finance / strategy analysts needing A/HK/US data, earnings pulls, spreadsheet reconciliation via chat.
- Research leads who want multi-site WebBridge sweeps → Excel/PPT without hiring an intern.
- Ops managers scheduling daily briefings via Cron on a dedicated always-on Mac Mini.
- Knowledge workers already on Kimi K3 who outgrew web chat context limits.
Weak fit:
- Software teams shipping production code — use Kimi Code, Claude Code, or jcode.
- Intel Mac shops — no native build advertised.
- Air-gapped compliance — cloud K3 inference + WebBridge still phone home for reasoning.
- Open-source mandate — OpenClaw or local K3 weights after July 27.
Install and first-run workflow
Download
| Platform | Link |
|---|---|
| macOS (Apple silicon) | kimi.com/products/kimi-work → Mac download |
| Windows | Same product page → Windows download |
Suggested first hour
- Create workspace folder:
mkdir -p ~/Documents/KimiWorkspace - Mount folder in Kimi Work settings — confirm Ask before acting is on.
- Install WebBridge extension → verify connected status in app.
- Run smoke prompt: "Summarize explainx.ai's latest Kimi K3 post and save bullet notes to KimiWorkspace/kimi-test.md" — validates file + model path.
- Optional Cron: schedule a 7 AM agent call with Keep Computer Awake for one week — measure credit burn on your tier.
- Swarm trial: on Allegretto+, run a 3-source competitor table → Excel job; compare time vs manual.
K3 selection: Kimi Work 3.1.0+ supports K3 per mobile/desktop K3 guide — confirm version in app About screen after install.
For a Chat-only coding anecdote on verification loops vs one-shot demos — same model family, different harness — see Kimi K3 vs Fable 5 vibe engineering (3D stadium E2E case study, July 20).
Summary
Kimi Work is Moonshot's desktop knowledge-work agent: local folders, WebBridge on real Chrome/Edge, Cron automation, K3 Agent Swarm into PowerPoint/Excel, and native A/HK/US finance data. It is not web Kimi chat — it is a scheduled, file-mounted, browser-driving digital employee with honest limits (Apple silicon Mac, sleep stops Cron, paid tiers for swarm volume, K3 subscription pause). Compare OpenClaw for open local gateways, jcode for repo-native coding swarms, and Cowork / ChatGPT desktop for Western-vendor policy paths.
Related on explainx.ai
- Kimi K3 vs Fable 5 — vibe engineering on a 3D stadium — Chat-harness anecdote: E2E verification vs fast single-file demo
- Kimi K3 API guide — 2.8T, pricing, 1M context
- Kimi K3 subscription pause — GPU capacity July 2026
- Kimi K3 mobile app and API guide
- Run Kimi K3 locally — open weights prep
- jcode agent harness — swarm and memory on desktop
- What Is OpenClaw? — local agent gateway primer
- OpenClaw Foundation — 501(c)(3) stewardship
- How to use Claude Cowork safely
- ChatGPT Work vs Codex — desktop quota context
- Cyber guardrails debate — K3 demand context
Official sources: Kimi Work product page · WebBridge features · Membership pricing · Kimi K3 tech blog · Kimi Platform API
Kimi Work features, pricing tiers, K3 subscription availability, and platform support accurate as of July 21, 2026. Verify download pages, membership signup status, and regional access before team rollouts — Moonshot paused new K3 subscriptions July 19 and reopens in batches.