Cowork with Agents in a Hive

Hive gives every coding task its own virtual machine and an AI agent to work in it. Nothing runs on your laptop. Start as many as you need and come back to work that’s waiting for your review.

You’re on the list. We’ll email you when your spot opens up.

Want to tell us a bit more? It helps us decide who to let in first.

AI shouldn’t need your computer to help you.

Most coding agents today run on your own machine, with your files, your credentials and your shell. That’s a lot of trust to hand to software that’s still learning.

We started Friendly AI so you can work with AI without giving it the run of your computer. In Hive, every agent gets its own virtual machine. It can install packages and break things, and none of it touches your laptop or your servers.

Friendly means the agent is on your side. It works somewhere it can’t hurt anything, it asks before it guesses, and you decide what ships.

Hive is the first thing we’re shipping. We’re letting a small group in first, and we’d like you to be in it.

The Friendly AI team

The whole job, not just the code.

Hive plans the work, builds it, checks it, ships it and keeps an eye on production. Agents do the legwork on machines in the cloud. You make the calls.

  • Our cloud or your machines

    Every task gets a fresh VM in Hive Cloud, or runs on a machine of yours, Mac or Linux, with the Hive agent installed. A Hive VM can even run desktop apps like Codex, so nothing has to run on your laptop.

  • Subagents, then whole workspaces

    Inside a session, the agent runs subagents that explore, build, test and review side by side. When one machine isn’t enough, the lead agent fans the job out to many workspaces, each with its own VM and checkout, and tracks them all from one chat.

  • See proof before you read code

    Each session ends with a verification page: what the agent planned, what it changed, and evidence it works, including a recorded walkthrough of the running app. Comment on any part and the agent picks it up.

  • Review, fix and merge in one place

    Pull requests and their checks live in Hive, sorted by what each one needs. A failing check or a merge conflict comes with a button that puts an agent on it.

  • Keeps going while you’re away

    Schedule work that repeats: watch open PRs every hour, triage new errors overnight, upgrade dependencies on Mondays. It all runs in the cloud with your laptop closed, and your phone tells you when something needs you.

  • Hear back from production

    Errors from your apps arrive in Hive with their stack traces. One click starts an agent on the fix and tracks it in Linear or GitHub.

Hybrid Input Virtual Environment

One shared space where people and agents work side by side. Work can start from a person, an agent or an event, and any agent can pick it up.

Work flows into Hive from people (you and your teammates), from agents (Claude Code and Codex) and from events (a failing build or a production alert). Hive turns each piece of work into a session on its own VM, where people and agents work on it together.

PeopleAgentsEventsYouYour teammateClaude CodeCodexFailing buildProduction alertHiveone shared spaceSessions, each on its own VM.People and agents work in them together. PeopleAgentsEventsYouYour teammateClaude CodeCodexFailing buildProduction alertHiveone shared spaceSessions, each on its own VM.People and agents work in them together.

Bring any agent

Hive runs Claude Code and Codex today, including the Codex desktop app on a Hive VM, and you can plug in an agent of your own. Each one works in its own VM, so none of them touches your machine.

Mix vendors in one job

Plan with one model, build with another and have a third review the result. You pick the agent for each step and can change your mind later without rewriting anything.

Join the waitlist.

You’re on the list. We’ll email you when your spot opens up.

Want to tell us a bit more? It helps us decide who to let in first.