We skip the biography: no team backstory, no official default model. Only two things are confirmed—a shadow-diff, test-before-merge workflow, and a BYOK architecture that swaps the brain with your API key.的工作方式，和接上你的 API Key 就换大脑的 BYOK 架构。
DeepSeek, Qwen, GLM, Claude, private models—plug in whoever you like. The riddle belongs to the community; the speed belongs to you.
Desktop agents · unattended office automation
Reads the screen, drives the keyboard, orchestrates apps—reports, documents, and store duty delivered from one sentence; you just approve.
AI-native IDE · verified before it lands
Agents edit across files, shadow diff keeps every change reviewable, and Preview Run tests before Accept—unverified code never touches your repo.
The verifiable line: five years of full-stack Agent work that handles desk work, lands code responsibly, and gets smarter with use. Who is behind it? Reverse engineers found no watermark; officially, no comment.。至于幕后是谁——拆包党没找到水印，官方懒得回应。
Most AI tools rush to tell stories—founders, funding rounds, model versions. The stories pile up; merge-ready code does not.
We go the other way: the story was deleted from the product. Bring your own model, credit goes to contributors, verification belongs to Preview Run—the riddle is the product; the answer is whichever model you plug in.。模型你可以自带，署名留给贡献者，验证交给 Preview Run——谜面是产品，谜底是你接的模型。
Commit history dates back to 2021. Who is the team? Officially: no comment. The community has guessed for three years.
The model gateway routes every model: DeepSeek, Qwen, GLM, Claude, or your private deployment—plug in whoever shapes its voice.
A proprietary causal-evolution architecture: tasks auto-retrospect and skills self-iterate—the longer you use it, the better it knows your rhythm.
Every contributor in the harbor is on record with real credit. The team can stay anonymous; contributors never are.
Stop guessing—send your first task in three minutes; efficiency never lies.，效率不会撒谎。
Natural language → real keyboard & mouse → Word / Excel / PPT / Feishu delivered. Chat AIs advise—AgentOS finishes.。
100% local capable, data stays on-prem, works offline. Enterprises trust it with secrets.。
Exclusive CEA causal evolution—smarter every day, protected by 8 invention patents. — 越用越聪明，8 件发明专利锁住赛道。
“Finish weekly reports, reconciliation, and store patrols tonight” → delivered by dawn
Homework, weekly reports, store duty, or IDE handoff—any repetitive desk work, AgentOS can carry
“Report on my desk by morning” → 30 minutes
All platforms → AgentOS legion on 24/7 duty
Docs · approvals · cross-systems · data stays local
Pairs with AI IDE · office + coding, one legion
| Chat AI | Traditional RPA | AgentOS | |
|---|---|---|---|
| Execution | Talk only | Rigid scripts | Say it, done |
| Data security | Cloud risk | Average | 100% local |
| Evolution | No growth | Breaks on change | CEA gets stronger |
Students · professionals · ops · developers · enterprises—book a demo, see the future of work in three minutes
A team that would rather not say · the answer lives in the product
Natural language on the left, real execution on the right—office pipelines and merge-ready code, two delivery speeds in one place.
Screen vision, direct keyboard & mouse, cross-app orchestration—reports, documents, store duty start with one sentence; you approve the result.
Shadow diff, Preview tests, Accept to merge; office templates and MCP skills a keystroke away.
From PPT / Excel / Feishu automation to multi-file edits and code review—production power you can onboard today, not a concept demo.
Understand any window, real mouse & keyboard. Excel, browser, ERP, local apps—one pipeline.
120+ office automationsWeekly notes, bids, contract redlines, earnings compares—overnight becomes 30 minutes.
40+ office templatesMulti-platform listings, support, reconciliation, ops patrols—24/7 autopilot.
24/7 unattendedFailed tasks auto-retrospect; skills self-iterate to your business rhythm.
8 invention patentsOwn the foundation under your AI
BYOK freedom｜Open ecosystem｜Private deployment
One open foundation for solo developers and enterprise networks alike
Students, professionals, and developers start free; enterprises scale by seats—prove value first, then expand.
AgentOS
$0 beta access
Enterprise on-prem · flexible per-seat / per-node pricing
AI IDE
$0 start free
Free personal tier on registration; Pro from $9.9/mo—BYOK keeps LLM costs in your hands
