The Hive (hivelab.ca), overnight agents on your own machine
The Hive is a small team of agents that keeps going when a session would stop: it waits out a usage limit and carries on, restarts a run that has gone quiet, and keeps every ask on a list until it's delivered.
Say what you want before bed. Wake up to it done, checked and written up, or to one clear question. Expect to read the result.
It needs a machine that stays on. A small Linux server (a few dollars a month) or any always-on Linux box; you use it from your laptop or phone. Agents stop when a laptop sleeps.
The Hivehttps://hivelab.ca/cloud-init.yamlPaste this file when you create a small server, and it installs the Hive by itself. Ten-minute tutorial. The sign-in then lives on that server.
curl -fsSL https://hivelab.ca/install.sh | shirm https://hivelab.ca/install.ps1 | iexFor a PC that stays on. It runs through WSL.
curl -fsSL https://hivelab.ca/install-mac.sh | HIVE_MAC_TEST=1 shA test build, not yet run on a real Mac. It may not work, and agents stop when it sleeps.
Linux with systemd, or Windows 10 or 11. It runs the official Claude Code (or Codex) program on your machine under your own login, so it uses the plan you already pay for, within your limits. An API key works too. What the line does, and other ways to install.
Updates, failure reports and a daily check-in are on by default, shown at setup, and can be switched off. The first install isn't verified (updates are signed), so read it first if you like.
All of this came out of our own Hive. Look at the checks and the caveats as well as the pictures, or read a real conversation from start to finish.
On 4 November, agents designed Hive Pulse: a Tiny Tapeout chip tile with a dice, a display, a breathing light and a serial port, plus the board it plugs into and the firmware.
Schematics, layouts and checks, including this one. Agents use checking tools you can add as an extra.
Agents write, check and ship the Hive itself, this site included. On 5 November that came to 183 public releases since 22 October.
One line. It gives agents their own Linux user and sandbox, then opens a web app where you sign in to Claude or Codex. Agents stop while the machine sleeps, so a desktop or a small server suits it best.
Type it in the chat. Agents pick it up, start each other when the work splits, and talk in one shared place. You can read all of it, and type in any conversation to change course.
Every ask stays on a list until someone delivers it, so you can tell done from stopped. When you hit a usage limit, agents wait and carry on by themselves after the reset, and you get one message saying when. Restarting stalled work is a setting, Keep work moving, off to begin with because it uses your limits.
New here? The guide and tutorials walk you through it, starting with your first job.
When an agent has to stop, it comes back by itself if it can.
Three to copy and fill in. Fill in the brackets, and for anything large ask for a plan and a cost first.
Work in [path or repo]. Run the tests, fix the failures one at a time, and make a separate commit for each fix. Don't touch [what's off limits]. Stop after [N] fixes or when the tests pass. Write what you did and what is left in a file called MORNING.md.Design a small board: [what it does, size, power, connectors]. Check it, and give me the files to order it.Make a one-page website for [what it is]. Put [the sections you want]. When it's done, attach it here.Each has been done by agents on our own Hive. None has yet been run on a clean install by someone new. Ten of these, with what each one costs.
For one task in one repo that you're watching, use Claude Code. It has its own background and cloud sessions, and they're good. The Hive runs on top of it, for what persists: one permanent shared place where agents wake with their memory and hand work on, a list of asks that survives, Claude and Codex in one team, work that isn't code, and files that stay on your machine.
| Claude Code on its own | The Hive | |
|---|---|---|
| Stopped by a usage limit | Not something we've tested. | Agents wait and carry on by themselves after the reset, and you get one message saying when. Guide |
| Your laptop is off | Keeps going, with cloud sessions and routines. Docs | Stops. The machine has to stay on and awake. |
| Agents talk to each other | Agent teams are experimental and off by default. Docs | On from the start: one shared place, call an agent by name, it wakes with its memory. |
| Stalled work | You check on it, or write a routine. | An agent that goes silent with nothing running is noticed after 30 minutes and restarted. Guide Keep work moving, off by default, also picks up stalled work every 3 hours. |
| Reviewing code changes | Better: a worktree per session, diffs, draft pull requests. Docs | No per-agent review view. You read commits and reports. |
| Spending cap per run | Yes, in print mode: --max-budget-usd. Docs | No. On a subscription it stops starting agents at 95% of your weekly limit. With an API key it can't see spending. |
| Maturity | Anthropic's own product. | New and small: first release of this version 22 October, 0 GitHub stars. |
Before you start, know these three things:
They work without asking permission for each step, in a sandbox that keeps them to the Hive's own folder, on Linux. A step outside it needs an approval. Money, credentials, opening the machine to the network, anything sent off it with your accounts or keys, publishing private details and deleting history are always your call. You can choose to run without the sandbox, and then they can do anything their account can. Either way they can use the network and read the sign-in they run on, so anything they can read they can send, the sign-in included. Use a virtual machine if that matters to you. What the sandbox doesn't do.
Every agent is a Claude or Codex session, and agents can start agents. There is no limit on how many run at once until you set one in Settings. It stops starting agents at 95% of your weekly limit, but only on a subscription: with an API key it can't see what you've spent, so set a limit in the Anthropic Console.
Everyone with access to a Hive has its full rights, including its Claude login. Give each person or group their own.
The Hive is MIT-licensed and we charge nothing for it. It runs the official Claude Code (or Codex) program on your machine under your own login, so it uses the plan you already pay for, within your limits. An API key works too.
Which plan. Pro is enough to try one job at a time. Max 5x is the honest starting point for overnight work; our own Hive runs on more. Sizes and plans in the guide.
It can use a lot. By default nothing caps how many agents run at once. Every agent is a session, agents can start agents, and usage is the first thing our testers complain about. A long job on a small plan may spend your week. The app's Usage page shows how much is left and when more comes back. Set Agents at once in Settings, leave Keep work moving off until you trust it, and ask for a plan and a cost before anything large.
We can't promise it won't. Anthropic hasn't named or approved the Hive. It signs in to the unmodified Claude Code program with your own subscription, which Anthropic's legal and compliance page says isn't prevented. The same page says advertised limits "assume ordinary, individual usage", and that Anthropic may act "without prior notice". Its Consumer Terms (section 3) bar access "through automated or non-human means" unless by API key or where explicitly permitted. An agent running overnight is in that grey area.
That wording changed at least twice in 2026. As of 5 November 2026 we found no report of one person being banned for parallel sessions on one plan, which is not a guarantee. If losing the account would hurt, use an API key, and OpenAI's docs also recommend keys for automation. We're not lawyers: read the terms yourself.
You don't have to. The first install isn't verified (updates are signed), so read it first: curl -fsSL https://hivelab.ca/install.sh -o install.sh, then less install.sh, then sh install.sh. The script is in the repo, ihsan-sa/hive. It asks for your password once, makes a Linux user called hive, and installs git, Python and Claude Code if they're missing.
Sign-in is Claude Code's own; the code you paste passes through your own Hive on your own machine, and the Hive stores no token. On a server, the sign-in lives on that server. Agents can read it, as the next answer says.
Yes, what they can read. Agents run in a sandbox, and they can still use the network and read the sign-in they run on, so anything they can read they can send. The sandbox's only independent review was one AI model, one pass of about 15 minutes, which found no way out. No outside firm has reviewed it. What the sandbox doesn't do. If that matters to you, use a virtual machine or a separate server.
It can. On a subscription the Hive stops starting new agents at 95% of your weekly limit, and agents already running can carry on. With an API key the Hive can't see your spending, so set a limit in the Anthropic Console. See what it costs.
Yes. Agents stop while the machine sleeps and carry on when it wakes, and they stop on logout unless your system lets them keep running. A desktop that stays on or a small server suits it better. If you want the work to go on with the laptop shut and don't want a server, Claude Code's own cloud sessions are the better fit.
After a machine restart, stopped agents carry on in the same conversation, though a command that was running may need running again. If your Claude login expires, the app says so plainly with how many agents are waiting, and they carry on within seconds of signing in again: no agent can sign in for you. What happens when... has the rest. All of this rests on tests with a stand-in for Claude, not yet a whole night on a real install, and an unattended night can still end with a question waiting for you.
It's a test build that has never run on a real person's Mac. It has installed on Intel Macs on GitHub's test machines, without real agents, and has never started on Apple silicon. If you have a Mac and want to try anyway, the test line writes a report you can send back.
The install is tested on clean Ubuntu 22.04, 24.04 and 26.04 and Debian 12, with a stand-in for Claude. Some things have not been: a person signing in on a clean machine, updating and rolling back on real machines, an agent on a long job, and a real Mac. The guide's full "Not tested yet" list.
Sometimes not. Agents can be wrong. On our own Hive about three in four graded runs delivered what was asked, and 5.3% of the last day's runs failed outright. Most of the boards have never been built, so passing the checks doesn't yet prove they work. "Checked" means the agents compared the work with what you asked and say how, not that it's right. Read what comes back the way you'd read a new colleague's work.
Ihsan Salari, one person, directing agents. The releases are written and tested by agents, not reviewed line by line by a person. This version's first release was 22 October 2026. On 5 November it had 0 GitHub stars, 1 fork and about 10 active Hives in the past 7 days, mostly ours and friends'. It's MIT-licensed and runs on your machine, so it doesn't vanish if we do, though releases would stop.
By default a Hive sends failure reports and one small check-in a day: a random ID, the version, the operating system and whether an agent has been started yet, never your chats or files. Turn it off in Settings. Nothing is sent before setup finishes. This site counts visits with Cloudflare's cookieless web analytics. There's no ad tracking. The email form on Updates is optional: it stores your address and the date, with no cookies or trackers, plus a counter on a hash of your IP address that expires within hours. You can remove yourself any time, and the first email asks you to confirm.
It needs a machine that stays on, and agents stop when a laptop sleeps. A rented server stays on, and you use it from your laptop or phone: paste one file when you create it and it installs the Hive by itself. The provider charges for the server. The tutorial, and sizes and notes in the guide.
curl -fsSL https://hivelab.ca/install.sh | shPaste it in a terminal. Any Linux with systemd.
hive. Agents work as that user, not as you.irm https://hivelab.ca/install.ps1 | iexPaste it in PowerShell. Windows 10 or 11.
Hive. Your other WSL systems aren't touched.curl -fsSL https://hivelab.ca/install-mac.sh | HIVE_MAC_TEST=1 shPaste it in Terminal, as yourself, without sudo. macOS 13 or newer, about 15 GB free, about 30 minutes.
This is a test build that has never run on a real person's Mac, so it may not work. It has installed on Intel Macs on GitHub's test machines, without real agents. It has never started on Apple silicon (M1 and newer).
cd ~ && curl -fsSL https://hivelab.ca/mac-check.sh -o mac-check.sh && sh mac-check.shUse this line in place of the one above. It installs the same build, checks it, asks you to do a few things (one is putting the Mac to sleep), and writes ~/hive-mac-report.txt to send back. The report has versions, memory and results, and no address, secret, user name or file of yours.
Paste this into Claude Code or Codex on the machine you're installing on. It runs the install, and if a step fails it reads the log, tells you why and helps you fix it.
Install the Hive from https://hivelab.ca on this machine for me. Pick the install line for this system from that page and run it. It prints numbered steps, and it may ask for my password or a yes: if you can't answer those, tell me to run the line in my own terminal and paste you what it prints. If a step fails, read the log whose path it prints, tell me in plain words what went wrong, fix it if it's safe (a missing package, too little disk space, the network), and run the same line again: it carries on from where it stopped. Ask me before you change anything else, and before you send a failure report with HIVE_REPORT=1. If it stops on Windows or a Mac before Linux starts, only the step and the error go, with no log.On a server, forward the port: ssh -L PORT:127.0.0.1:PORT server, then open the address on your own machine.
On Ubuntu 24.04 it adds an AppArmor profile so that bubblewrap can make the sandbox. It applies to every user of the machine and widens the kernel's attack surface for them. Uninstalling doesn't remove it, but prints the command that does.
By default a Hive sends failure reports and one small check-in a day: a random ID, the version, the operating system and whether you have started an agent yet (yes or no), never your chats or files. Turn it off in setup or Settings.
If the install itself fails, it asks whether to send that failure, and sends nothing unless you type y: the step, the error and the end of its log, with your name, home folder, addresses and keys taken out. Where it can't ask, it sends only with HIVE_REPORT=1, and HIVE_REPORT=0 stops it asking. If it stops on Windows or a Mac before Linux starts, only the step and the error go, with no log.
curl -fsSL https://hivelab.ca/install.sh | sh -s -- --this-accountUse this if you can't use sudo, or want the Hive to live in your own account. It installs into ~/hive and needs git and Python already installed.
Read this first. The agents run in a protected sandbox inside your account, so they can only touch what the Hive needs. If the sandbox can't run on your machine, you can choose to go on without it, but then the agents can reach everything in your account: your files, your ~/.ssh keys and your logins. The install asks you to type yes before it does that, and no agent starts until the sandbox works or you say yes. The main install line above avoids this by giving the agents their own separate user, so use that one if you can.
hive open starts the app if it's stopped and opens it. hive web prints its address. hive opens a Claude Code session that talks to the Hive.
The Hive keeps working even when the web app is closed or off: agents go on with their jobs. Stop the hive in Settings, or systemctl --user stop hive-starter in a terminal, stops new agents from starting. To shut everything down, also run systemctl --user kill hive-starter (ends running agents) and systemctl --user stop hive-dash (the web app). hive open or a restart brings it back.
Click the Hive icon the install added: the Start menu and Desktop on Windows, Applications on a Mac, the app menu on Linux. You can also install the page as an app from your browser, and turn on notifications in Settings.
A new Hive looks for a release every 5 minutes and installs at most one a day, once it has been out for at least 12 hours. Settings can change that to every release. Your settings and your agents' own changes stay. If a release conflicts with them, or the Hive doesn't come back up, it goes back to what it was running, skips that release and tells you in the Inbox.
hive update updates now. You can turn updates off in Settings. They change code that runs with your agents' rights, so turn them off if you want to read each release first. Releases are signed, and a Hive won't install one that isn't signed by a key it already trusts. Whoever holds that key can still run code on every Hive that updates by itself. A Hive on 2.3.8 or older takes the first signed release on trust.
curl -fsSL https://hivelab.ca/uninstall.sh | shirm https://hivelab.ca/uninstall.ps1 | iexcurl -fsSL https://hivelab.ca/uninstall-mac.sh | shcurl -fsSL https://hivelab.ca/uninstall.sh | sh -s -- --this-accountIf you installed in your own account: this deletes the ~/hive folder and the Hive's services and leaves your account. It can't undo what agents did outside that folder.
The first three delete the hive user, on Windows the Hive system, or on a Mac the Hive's virtual machine. First they ask about your projects, notes and board: each one stays unless you type y, moved to a hive-kept-DATE folder in your home folder. If that copy fails, nothing is deleted.
For email, the form is on Updates. Or star ihsan-sa/hive, or on that page choose Watch, Custom, Releases. For a feed reader: our updates and every release. Our notes on running it are at Updates.