Everyone gave their agent a computer.Only one of them is yours.
OpenAI, SpaceXAI and Meta have all launched always-on agents since August. Each one runs only on its maker’s own model. Machine gives you the computer and lets you choose the rest, the model included.
Any model · Lives in your repo · Answers to you
Six things you want from an agent. Machine does all six.
01Runs on any model — Claude, GPT, Gemini, Grok, or your own key.
02Keeps its skills and memory in your repo, as files you own.
03You decide which actions need a person’s approval.
04Talks to you in Slack, on the web, on desktop and on your phone.
05Can use your own computer, not just one in the cloud.
06Built for your whole team, not one person.
Machine does all six. Each of the others runs on one model only.
Start with MachineAsk for something. A computer boots.
A machine comes up
Its own isolated Linux machine: its own filesystem, processes and network. Nothing shared with another session.
Your repo clones
Agents, skills, memory and triggers arrive with it, because all of them are files in that repo.
A branch is cut
Every edit, commit and stray file the session produces lives on that branch and nowhere else.
You review the change
Work reaches main through a change request. Read it in Slack or on the web, then approve it or send it back.
Dots, Grok Bot and Muse all give the agent a cloud computer too. None of the three has announced putting that work on a branch in your own repo for you to review as a change request.
It has a whole computer. You hold the rules.
Its own machine, every session
One isolated cloud computer per session. It can install anything and break anything without touching another session, and what it commits lands in your repo as work you can review.
Allow, ask, or block — per action
For any tool the agent can call, you set a rule: always run, ask a person first, or block. Rules are files in your repo, so they are reviewed like everything else.
Connector credentials never enter the machine
Slack, GitHub, Notion and the rest are reached through one scoped token brokered outside the machine. The agent gets the access, not the keys.
Approval rules are off until you set them. That is deliberate: a project with no rules runs everything, and a project with rules runs exactly what you said.
Same money. Very different month.
ChatGPT Pro 100 is $100 a month. Here is what that money does on Machine.
ChatGPT Pro 100
$100/ mo
- A dot included
- One model: GPT-6 Astra
- In ChatGPT, Codex, Slack and Teams
- Individual plan; teams via Business Premium
Machine
$80/ mo$20 left over
- 2 seats
- 5,000 pooled credits a month
- Every model, or your own key
- Agents live in your repo
Machine: $40 a seat a month with 2,500 pooled credits, or free with 200 credits. Rival prices as published or reported on September 30, 2026.
Keep the plan you have.
Sign in with your ChatGPT or Claude subscription and Machine runs on it. Or paste any model key. Either way, the agent, the repo and the review are still yours.
- Your ChatGPT plan
- Your Claude plan
- Any API key
Where they are ahead today.
OpenAI
Dots
- Reaches 4,000+ apps through OpenAI’s plugin ecosystem
- Specialist Dots with their own identities and credentials
- Lives in Microsoft Teams as well as Slack
SpaceXAI
Grok Bot
- Watch it once, and it saves what you did as a routine
- Bots message each other and work in parallel
- Signs into tools with no API at all
Meta
Muse
- Shops and pays through Walmart, Sephora and Shopify, with Expedia coming
- A live video avatar, announced for soon
- Coming to Meta’s smart glasses, hands-free
Machine is not trying to be a shopping assistant or a video avatar. It is the one built for work you can review, on a model you choose.
Things people ask before switching.
Own the agent.Keep the work.
Start free with 200 credits a month for sandbox compute, and bring the model you already use.