Free
- 1 box - 60 minutes to build, claim it by email to keep it
- 1 vCPU / 2 GB RAM / 10 GB disk
- Sleeps after 24 hours idle
- Free AI: 10M Human.01 AI tokens a month, plus free models
- 6 images an hour, 20 a day
Press Launch and get your own computer in the cloud in seconds, with an AI agent that builds for you and a Bot that drafts replies to your chats. 10 million AI tokens a month are free, with no account, no credit card and nothing to install.
Already have a box? Log in to open it
Other ways to sign in (Google, wallet) Back to the start page
Reading…
Plug the phone into this computer. The box talks to it over the cable, through this page. The phone never leaves your desk.
Turn on USB debugging on the phone first (Settings, Developer options).
Catches every app that uses the system proxy. Apps with their own network stack (some messengers) ignore it and need a rooted device.
Open this box from the terminal on your own computer. Paste this there, once:
After that, ssh vm.human01.xyz opens this same box from that computer. The command works once, within 10 minutes.
Connect this box to GitHub so you and your AI agent can clone, push and open pull requests. Takes about a minute, and no password is typed here.
Let us make your free GitHub account.
....-....This page notices by itself when you are done. The code works for 15 minutes.
Connected as . git and gh are ready in this box.
Keep your work safe on GitHub
gh repo clone..env, keys and node_modules are never uploaded.Auto-save is on. Your project is saved to
How it works
git in the Shell tab. The connection stays.Bring a repository into this box
Your GitHub connection follows your claimed box: it is ready in every box of yours, including new ones. Clone a repository into the app folder, then open it in Files or tell the agent about it.
One quick check that you are a person, then your box opens.
Claim it with your email and everything stays exactly as you left it.
| No. | Time | Source | Destination | Protocol | Length | Info |
|---|
Ask what a customer would ask. The AI answers only from your knowledge folder, the same way it will on your website.
It answers only from the files in the box's knowledge folder: your documents, the FAQ, your rules, pages of your website and past chats. Change a file and it reads it again by itself.
Paste this code once, just before </body> on every page. Your website can be hosted anywhere; the chat talks to this box.
Answers come from your knowledge graph and cite their sources.
Add this box to your own coding agent so it can look things up in your knowledge. It adds one MCP server, named after this box - knowledge - with three tools: one finds passages and facts with the page they come from, one shows you the original figure, table or page itself, and one follows how ideas connect. Add as many boxes as you like; each keeps its own name. Pick your tool:
A picture costs your agent far more than a passage. If you would rather keep every turn small, change /mcp to /mcp-text in the command above: same key, text and citations only. The key is secret - anyone who has it can query this box. Regenerate it if it leaks (your harness will need the new command). Rename the box and this server name changes with it: copy the command again and remove the old entry.
Add a file or a link and its ideas light up here.
Drop to add to your knowledge
Any terminal on Mac, Linux or Windows. No account, no credit card.
$ ssh vm.human01.xyz
No SSH key yet? Run ssh-keygen -t ed25519 and connect again.
Your own Linux computer opens in seconds. Type agent and tell it what to make: 10 million AI tokens a month are free.
A free box gives you 60 minutes to build. Claim it with your email and it stays yours; a box nobody claims is deleted after 24 hours.
●
Choose who you are, then press Try this. Your box opens with the first step already written for you, and you press Start.
Every box is a hardware-isolated virtual machine with its own Linux kernel - not a shared container. Break things freely.
One virtual core and 2 GB of RAM on Free, enough to build, test and run real projects. Paid plans get 2 vCPU and 4 GB.
A private home directory that persists for as long as your box lives.
From pressing enter to a prompt. Boxes boot on demand, no queue and no signup.
Debian 13 with the essentials installed, plus an AI coding agent: type agent to start opencode, already connected to Human.01 AI.
Launch opens your box right in this page: the agent, a terminal, your files and the Bot, one tap apart. It works the same on a phone.
The AI agent (opencode) opens first. Switch to a normal terminal, your files or the Bot, and the agent keeps working.
Browse your project, edit code in a real editor, upload files or drag and drop them in. Paste an image, audio or file straight from your clipboard.
Connection, time left, images left, whether your app is running, git, RAM and load, and disk, always at the bottom.
Run your app on $PORT and open it at its own address. It is private to you until you claim the box.
Connect GitHub in about a minute with a device code, with help to sign up if you have no account. Every change is then saved to a private repo about a minute after you stop.
Files such as .env, keys and node_modules are never uploaded.
A one-time pairing code links your laptop's ssh key to this same box. Then ssh vm.human01.xyz opens it from there.
The workspace works on a phone screen, and you can install it as an app.
Pick one at the top of the page. The terminal, the claim email and the Bot follow your choice.
Type agent and start building with opencode, already connected to Human.01 AI. No API keys, no setup, nothing to install.
Every account gets 10,000,000 tokens of Human.01 AI a month, free, metered the way Claude and Codex do it: a 5-hour session and a rolling week, each with its own cap, shared by all your boxes. Pro gets 60,000,000.
Qwen3.8 Flash Next 180B runs on our own A100 GPUs and works with up to 128k tokens of your chat and code at once.
Fair use keeps it quick for everyone: up to 30 requests a minute, 3 at a time.
From OpenCode, Kilo Code and OVHcloud. They are shared and rate limited, and they don't count against your 10M a month.
opencode reads, edits and runs code inside your box. Describe what you want, watch it work, and keep the result.
The agent has a search tool that finds the right code in your project quickly, instead of reading every file. Very large projects fall back to grep.
Ask for a hero image, illustration or background and the agent saves a PNG straight into your project, made by Qwen-Image on our own GPUs.
A built-in frontend design skill pushes the agent towards distinct typography, colour and layout instead of the same template every time.
The gauge button in the workspace opens your AI usage: how much of your 5-hour session and of your week is used and when each resets, updated live while the agent works.
Session and week as two bars, today's tokens broken down by model, and the last calls the agent made.
It spots wasted context, such as the same part sent twice or a part that is far too big. Each finding has a button that copies a message for the agent.
The watcher never changes a request. It only observes, and you can switch it off.
Human.01 AI is the default. The free models below are one keystroke away: press ctrl+p in the agent and pick a model.
The default. Runs on our own A100 GPUs with a 128k context, and every account gets 10,000,000 tokens a month free.
Three fast models, offered free by OpenCode. None of them train on your prompts. DeepSeek and GLM keep no prompts; Beam keeps them for 30 days for safety.
Seven free models, including very long context.
Nine open-weight models hosted in the EU.
The free models come from outside providers. Everyone using Human.01 boxes shares their rate limits, so they can be slow or busy at peak times, and the providers may log prompts. Keep secrets out of them.
No commands to learn. Type what you want in plain words, and use the shortcuts below. Pick any example to watch it.
The Bot is an assistant that reads your chat apps on its own computer screen in your box and drafts the replies. You check each one and press Send.
The Bot works in a real browser inside your box. Watch it live, and take control whenever you want.
WhatsApp works directly. Telegram, WeChat, LINE and Messenger are read from their own web pages, with a screen reader as fallback. You sign in with the app's QR code on the Bot's screen.
It writes a draft for each reply and waits for you. Automatic sending is off unless you switch it on.
Payments, codes and private matters always wait for you. Add your own rules in plain words.
Do a task once while it watches, and it can repeat it, for example every morning.
Set a job for a number of minutes from now, daily, on weekdays or weekly. Its memory lasts, and your own documents become its library.
Send a command, a photo or a file to your own WhatsApp chat (Message yourself). It does the job and answers there, and can even write and run code in your box to get it done.
A log of what it did, its drafts and its stats, so nothing happens out of sight.
Every box can try the Bot with 20 sent replies, free.
At most 10 replies are sent in any hour. The Bot uses your box's daily AI allowance.
A LAB box trades the Bot for a reverse engineering workbench. Ask about an app and the agent inspects the real thing, then shows you the evidence behind every answer.
Electron and .NET apps, Go and Rust binaries, websites and saved captures. The agent reads the code and explains the behaviour it finds.
Ghidra, JADX and apktool come installed. No licence to buy, no setup: the agent drives them and reports back with the code it read.
Plug it into your computer and the box can reach it. Pull an app off the phone, take it apart, then build a new version and put it back.
Watch requests pass through your box as they happen, then hand the whole capture to the agent and ask it to explain the protocol.
Compiling and signing an app runs on GitHub Actions, not in your box, so the box stays quick and the build gets a real machine.
Every finding names the file and the line it came from, and says plainly what it could not determine.
LAB is for lawful research, education, interoperability, software preservation, and analysis of software you own or are authorised to examine.
Each film follows one business and the job its box does for it. Watch how it works, then try the same in your own box.
"Can my cameras tell me when a shelf runs empty?" Their box set it up in one ask. The same cameras can also watch the back door, the queue, the chiller and more, from one sentence each. Works on the free plan too.
Describe the part and the agent designs it in real millimetres. The box slices it and sends it to your own printer.
Each one is a film today and a box next. Join a waitlist and you are first in when it opens.
The agent writes the firmware, flashes your board over USB, and watches the signal on your scope with you.
For coding teachers: every student's editor on one wall, the stuck ones flagged, and you step in to point and talk.
Every plan includes the AI agent. Free boxes pause after a day idle; paid plans keep your box running with more room and more AI. Prices are per month in US dollars.
Paid plans add premium models from the Human.01 gateway, billed from your monthly AI credit. Free keeps Human.01 AI and the free models above.
No credit. Human.01 AI and 19 free models from OpenCode, Kilo Code and OVHcloud.
Everything in Free, plus every premium model below from a small monthly credit allowance.
The same models with more monthly credit and a priority queue, for heavier use.
Our own routed models: pick fast for quick answers or strong for hard problems.
More than 50 models are in the gateway catalog, and the list changes over time. Premium models are billed from your AI credit at published per-token rates. AI credits launch during the beta.
Open your profile, pick a plan, and send the exact amount shown to our own deposit wallet from your wallet app. No payment provider in between. Longer plans cost less per month.
Pay on your mobile carrier bill: no card, no code to type. Carrier billing is being set up with our first telecom partner.
Plans are in a private beta. The always-on boxes are available to beta members now; AI credits for premium gateway models and carrier billing arrive during the beta. Prices may change before the public launch.
Your box is free to use right away. When the hour is up it pauses; claim it with your email and it's yours to keep. Don't, and it disappears.
Full access. Install, build and run whatever you like.
The box is paused. Open the claim link from your welcome message and confirm your email to keep everything exactly as you left it.
Unclaimed boxes and their files are removed.
Boxes are generous but not unlimited. These rules keep the service open for everyone.
No. Press Launch and this browser remembers your box, so coming back opens the same machine. Over SSH, your public key identifies it: the same key lands in the same box. To use one box from both, press "Use my own terminal" in the workspace.
The box pauses and your session ends. Your files are kept while it waits to be claimed. Connecting again shows how long is left to claim it.
Your welcome message includes a claim link. Enter your email, confirm it from the message we send, and the box becomes yours, with your files intact. A claimed free box sleeps after 24 hours idle and is deleted after 7 idle days.
A virtual machine. Each box runs its own Linux kernel on a hardware-isolated hypervisor, so what happens inside stays inside.
Yes. Run your app on $PORT (8080) and open https://<boxid>.vm.human01.xyz - your box id and address are in the welcome message. Until you claim the box only your own connection can open it; once claimed it is public. Nothing else can connect in, and outbound traffic is limited to web and git.
No cost. The default is Human.01 AI (Qwen3.8 Flash Next 180B) on our own GPUs, plus 19 free models from OpenCode (DeepSeek V4.1 Flash, GLM-5.3-Flash and Beam, free for a limited time), Kilo Code and OVHcloud that you can switch to inside the agent. The free ones are shared and rate limited. Paid plans add premium Claude, GPT and Gemini models from a monthly AI credit.
Every account gets 10,000,000 tokens a month of Human.01 AI (Qwen3.8 Flash Next 180B on our own A100 GPUs), metered the way Claude and Codex do it: a 5-hour session of 350,000 tokens and a rolling week of 2,400,000, each resetting on its own, with up to 30 requests a minute and 3 at a time per box. All your boxes share the two windows, and your first session after claiming is doubled. Pro accounts get 60,000,000 a month. The 19 free third-party models come on top and don't count against it. The AI usage panel in the workspace shows both bars, live.
The Bot is an assistant that answers your chat apps from its own screen in your box. It uses apps you sign in to there yourself, with their QR code. It writes drafts and you press Send: nothing is sent without you unless you switch automatic sending on. Payments, codes and private matters always wait for you, and its log shows everything it did.
With crypto, straight to our own deposit wallet: open your profile, pick a plan, and send the exact amount shown from your wallet app. Or redeem a prepaid voucher: redeem it from your terminal and the plan starts. Carrier billing, where the charge appears on your phone bill, is coming soon. Plans are in private beta, so join the beta to get access.
Ask the agent for an image and it creates a PNG in your project with Qwen-Image. Each box gets 6 images an hour and 20 a day, and prompts are logged to catch abuse.
English, 简体中文, 繁體中文 and Bahasa Melayu. Pick one in the menu at the top of the page. The terminal, the claim email and the Bot follow it too.
Any terminal on Mac, Linux or Windows. No account, no credit card.
$ ssh vm.human01.xyz