Auto, when you are ready
With Auto on, a coworker carries a goal forward in the background, within a budget, and only reports done with evidence. Off, it waits for you.
MITOpen source · Self-hosted · 14 AI providers
Your AI workforce. Never miss the best AI.
Javis OS turns Claude, ChatGPT, Gemini, Grok and the models you choose into one 24/7 workforce on your own machine. Each coworker has a role, shares the same memory and tools, and keeps working in the background. When a better model ships, switch and keep everything they learned.
Free and open source
Use the AI plan you already pay for
Windows, macOS, Linux, Docker
Hi, I'm Javis!
ResearcherFinding sources
WriterDraft saved to Wiki
PR mergedChecks passed, site live
One workspace for the brains you already use
Claude CodeChatGPT (Codex)Grok BuildAntigravity CLIOpenRouterOpenAI APIAnthropic APIGoogle GeminiGroqDeepSeekOllama CloudOllama localOpenAI-compatibleAnthropic-compatibleClaude CodeChatGPT (Codex)Grok BuildAntigravity CLIOpenRouterOpenAI APIAnthropic APIGoogle GeminiGroqDeepSeekOllama CloudOllama localOpenAI-compatibleAnthropic-compatible
Sound familiar?
Your competitors already have an AI workforce.
Your competitors already have an AI workforce.
Do you?
AI changes every week. You pay for several, yet you are still the one switching tabs, re-explaining the project and pasting answers back and forth. One model per tab, one task per place.
- 01Falling behindOthers already have AI working for them around the clock, while you still do the work by hand, one chat at a time.
- 02Betting on the wrong AIA different model leads every few weeks. Commit to one and you miss the next, or juggle them all and carry the context yourself.
- 03Lots of AI, little doneAnswers stay buried in chats instead of becoming finished work and knowledge that grows.
Claude×
ChatGPT×
Gemini×
Grok×
Sheets× notes-final-v3.txt×
Javis OS shared memory
One memory for every modelMEMORY.md
Coworkers hand off the work4
Results saved as your files.md
Watch
See Javis OS at work.
From many models to one workforce
One request.
One request.
A whole team on it.
Give each coworker a role and the model that fits it. Javis passes the work along, keeps the context and saves the result where you can find it again.
1
Ask in plain wordsBy chat, by voice, or from Telegram, Slack and more.
2
The right coworker takes each stepResearch, writing and review, each on the model you picked.
3
Results land in your brainFiles you keep, memory that grows, work you can build on.
Illustration of a typical workflow. You choose the model for each coworker.
Your team
ResearcherGemini
WriterClaude
ReviewerChatGPT
PublisherGrok
Coworkers
Hire coworkers,
Hire coworkers,
not chat windows.
Each coworker has a name, a role, its own model and its own memory. Write the job once, give it the skills it needs, and switch on Auto when you trust it to carry work on its own.
1Auto on or off
2A model for each coworker
3Role, prompt and skills
The right model for each role
Claude for writing, ChatGPT for review, a cheaper model for chores. Change it any time; the role and memory stay.
A job description you can read
Role, instructions and skills live as plain files in your brain, so you can read, edit, share or import them.
Teams
Put them in one room.
Put them in one room.
A lead runs the work.
Group coworkers into a team with a lead who splits the brief, hands each part to the right person and checks the result. You talk to the room the way you would talk to people.
1A lead coordinates
2They reply to each other
3Talk to the whole team
A lead who delegates
The lead reads the brief, picks only the coworkers it needs and accepts or sends back their work.
A shared room, a shared history
Everyone sees the same thread, quotes each other and builds on the last answer instead of starting over.
Share a team in one file
Export a team as a .team.javis.zip with its coworkers, skills and pinned guidance. Keys and chat history never leave.
Multi-channel chatbot
One inbox for every channel.
One inbox for every channel.
Answers while you sleep.
Turn a coworker into a chatbot for customers on WhatsApp, Messenger, Instagram, Telegram, Slack, Lark, Zalo and more. Every conversation lands in one inbox, and you can take over any chat with one click.
WhatsAppMessengerInstagramTelegram
Slack
LarkThreads
ZaloZalo OA
SlackLarkThreadsZaloZalo OA
1How much the AI handled
2First reply in seconds
3On duty at night
One inbox, every channel
Customer chats from each connected account arrive in one place, with the AI reply and the person who sent it.
You stay in control
Set what the bot may do, see the chats where it stayed quiet, and take over a conversation whenever you want.
Numbers that show the value
See how many messages the AI answered, how fast it replied and how many came in after hours.
Real screens from the founder's own Javis OS, shown in English. Other people's names and data were removed.
Never miss the best AI model
Rent the model.
Rent the model.
Own the brain.
A new model takes the lead? Switch on the Models page and keep going. Your memory, skills, coworkers and connections stay exactly where they are.
Memory kept
Skills kept
Coworkers kept
Connections kept
Second Brain
MCP Hub
Skills
Tasks
Schedules
Coworkers
Claude
What makes it a workforce
More than a chat box.
More than a chat box.
A workspace that keeps working.
A Second Brain that grows every day
Memory, Wiki and Sources live in a markdown vault you can open in Obsidian. After each conversation Javis distils what it learned, and every learning pass is a git commit you can undo.
1,909 notes2,534 linksThe founder's own brain
Coworkers with a purpose
Specialists with their own instructions, memory and model, chained into workflows with verification.
Works while you sleep
Hand over a goal. The AI writes the spec, picks a worker and reports back when it is done.
Queued
Running
Done
Weekly report
Reply drafts
From an idea to a pull request
Connect a GitHub repo and ask in chat. Javis works on its own branch, opens the PR, watches the checks and fixes what fails. This very website moved to javisos.com that way.
Talk where work happens
Hands-free voice on the dashboard, plus the same Javis on Telegram, Slack, WhatsApp, Zalo and a CLI. Replies come in the language you write in.
TelegramSlackWhatsAppZaloCLI
SlackWhatsAppZaloCLILoops and reminders
Jobs on an interval, a clock time or cron, each checking its own work.
Every day · 07:00Every 30 min0 9 * * MON
Skills and plugins
Drop in a folder to teach know-how, or add a native tool every model can call.
skills/weekly-report/SKILL.md
agents/researcher.md
workflows/blog-post.md
You set the limits
Three permission levels, enforced by Javis itself, for every connection and job.
The real workspace
See the work.
See the work.
Keep the knowledge.
Real screens from Javis OS. Desktop shots come from the founder's own workspace.
Connected to your world
Plug in the tools
Plug in the tools
you already use.
Connections run through MCP, with several accounts per service. Every model gets the same toolbox, so switching brains never breaks a workflow.
Gmail
Google CalendarGoogle SheetsSlackZaloLark
Shopify
Pancake POS
Cloudflare
n8nAny MCP serverGmailGoogle CalendarGoogle SheetsSlackZaloLarkShopifyPancake POSCloudflaren8nAny MCP server
Meta
Meta Ads
Google Ads
TikTok Ads
X
Substack
Webcake
Botcake
systeme.io
Hostinger
HiggsfieldMetaMeta AdsGoogle AdsTikTok AdsXSubstackWebcakeBotcakesysteme.ioHostingerHiggsfield0AI providers to choose from
0permission levels you control
0languages in the docs
0%Javis cloud between you and your model
Built around ownership
Your machine.
Your machine.
Your files.
Your rules.
- Readable without JavisMemory, skills, coworkers and workflows are plain markdown. Open them in any editor.
- Versioned and portableSync the whole brain to your own private GitHub repo, shared between laptop and server.
- You know where data goesA request goes only to the provider you picked. With a local Ollama model it never leaves your machine.
Start with your first coworker
Make room for your AI workforce.
Install Javis OS, sign in to a brain on the Models page, and give it one real task. Then open http://localhost:7777 and meet your workforce.
Download for WindowsWindows 10/11 · installs Git, Python, Node and the AI brains for you
PS>
git clone https://github.com/blogminhquy/javis-os.git javis; cd javis; powershell -ExecutionPolicy Bypass -File install.ps1Download for macOSmacOS 11+ · open it like any app
$
git clone https://github.com/blogminhquy/javis-os.git javis && cd javis && chmod +x install.sh && ./install.sh$
git clone https://github.com/blogminhquy/javis-os.git javis && cd javis && chmod +x install.sh && ./install.sh
Get a Hostinger VPS with an extra 20% offAn extra 20% off when you sign up through the Javis OS referral link · Javis OS is an official Hostinger Partner
URL
https://raw.githubusercontent.com/blogminhquy/javis-os/main/docker-compose.hostinger.ymlIn hPanel, open your VPS, go to Docker Manager, choose Compose, select URL, paste the Compose link above and press Deploy.
$
curl -fsSLO https://raw.githubusercontent.com/blogminhquy/javis-os/main/docker-compose.yml && docker compose up -dEasiest way: give the repo link to Claude Code or Codex on your computer and say “install Javis OS for me”.
01
Install
Use the installer or one command. Javis opens at localhost:7777.
02
Connect a brain
Sign in to Claude, ChatGPT, Grok or Antigravity, or paste an API key on the Models page.
03
Give it a real task
Ask for an outcome you can check. Then teach it, connect tools and grow the team.
Good to know
Before you begin.
Is Javis OS free?
Yes. Javis OS is open source under the MIT license. You pay only your AI provider, through a plan you already have or an API key, and your own server if you use one.
Do I need to know how to code?
No. The Windows and macOS installers set everything up, and models are connected from the dashboard. Developers can go further with GitHub, plugins and their own MCP servers.
Can I mix different models?
Yes. Use a strong model for conversation, a cheaper one for background work and a local model for private notes, all on the same brain.
Where is my data stored?
On the machine or server where you run Javis, as plain files. When a model answers, the request goes to the provider you chose for it. There is no Javis cloud in between.
Can I run background work on a subscription plan?
Subscription plans are meant for personal use. Round-the-clock background jobs may fall outside that, so for heavy background work, point it at an API key or another provider on the Models page.
Built in the open by Nguyễn Minh Quý
Javis OS is an independent open-source project. Read the code, follow every release in the changelog, and help shape where it goes next.
Community
Ask, share and learn
Ask, share and learn
with the Javis OS community.
Meet other Javis OS users, get help when you are stuck and see what people are building.
Hire every AI.
Keep everything they learn.
Your models. Your coworkers. One workforce that remembers.
From the blog
Learn Javis OS,
Learn Javis OS,
one guide at a time.
Step-by-step guides on installing Javis, choosing a brain, running background work and more.
Connect Javis OS to Telegram: ask questions and get reports on your phone
Create your own Telegram bot in 5 minutes and message Javis like a person: ask for figures, send images, give voice commands, and receive background task reports without opening the dashboard.Read article
Assign background work to AI: Tasks, Reminders, and Loops in Javis OS
Three ways to keep Javis OS working when you are away from your computer: assign a one-time task, set a timed reminder, or create a loop that runs on its own. When to use each, and how to run them safely.Read article