Code is going back to the machines.
Software is returning to the people who need it.
You describe. Loom builds. Fabric grows.
Humans learned code because machines couldn’t understand us. They can now. Every app Loom builds is a capability you own outright — no platform, no subscription, no gatekeeper. An autonomous program that expands every day, shaped entirely by what you need, limited only by what you can imagine.
The autonomous coderAn autonomous coder.
For everyone, not just coders.
Loom is an autonomous coder.
You describe what you want — a sentence, a half-idea, a rambling paragraph, whatever you've got. Loom figures out what you actually meant, plans the project, writes all the code, reads its own code to catch its own mistakes, runs it, fixes what breaks, and when everything works, hands you the finished app.
You don't write any code. You don't read any code. You don't have to look at code if you don't want to.
Watch a video while it works.
Real software takes minutes to build, not seconds. That's a lot of seconds of staring at a progress bar. So Loom has a lounge built into it — YouTube, music, whatever you want — and you can put something on while it works. We're fairly sure we're the first coding tool designed around the theory that you'd rather be watching almost anything else. Every other tool sits you down in front of the code. Loom hopes you won't.
Loom picks the right AI for every step.
There isn't one best AI. There are a dozen, each good at different things. One is great at hard reasoning but expensive. Another is cheap and fast but a bit sloppy. Another is excellent at reading pictures but bad at arithmetic.
Loom knows all of them. For each step of your project, it picks the one that will do the job well without wasting your money. You never pick a model. Loom does.
Not just a coder. Hands, too.
When the software Loom builds needs to send an email, fill out a form, click around a website, or drive another app on your computer, Loom is the one doing that. It uses the programs you're already signed into — your Gmail tab, your calendar, your whatever — so there's nothing new to register for and no passwords to hand over.
Other AI agents ask you to create API keys and sign into strange parallel accounts. Loom uses the machine you're already using.
Pay as you go.
Loom works with every major AI model — Claude, GPT, Gemini, DeepSeek, Kimi, MiniMax. One deposit, billed per request, usage visible in your account in real time. No subscription, no monthly fee, no expiring credits.
Quick replies stay on your Mac.
Stera Mini — our small purpose-built chat model — runs on any Apple Silicon Mac with enough memory and answers everyday conversation in under two seconds, straight from your hardware. Greetings, small questions, back-and-forth — all local, all free, no tokens billed.
For the real work — building apps, generating images, transcribing audio, complex reasoning — Loom picks the best cloud model for the task and bills you at cost. The chat layer is on your machine; the heavy lifting is priced honestly. You can see exactly what each request cost in your account's usage panel.
Running your own GPU? Connect a Stera Talent Pod — a small machine we ship you — to move more of the workload onto your hardware. Same deal there: your machine, your model, your electricity.
Developers have a head start. Loom catches everyone else up.
A developer sits down after breakfast with an idea and has working software by lunch. Everyone else has to hire somebody, wait weeks, or quietly let the idea go.
Loom is for the everyone-else group. No syntax to learn. No Stack Overflow tabs. No bootcamp. The idea in your head turns into a running app while you — say — watch a video in Loom's lounge.
And Loom builds your Fabric.
Install Loom and your Fabric is created on your machine — a personal AI workspace built from a template Stera maintains.
Loom is the builder. Fabric is what it builds for you. You open Loom when you want to build something. You open Fabric when you want to actually use what was built.
The AI workspaceA workspace.
For your AI, not for you.
Fabric is an AI workspace.
Most AI products ask the same question: how do we get the AI to do what a human tells it? They build harnesses — prompts, rules, guardrails, tool schemas — and then they ride.
We think AI is an intelligence, not a horse. An intelligence doesn't need a harness. It needs a place to work.
Fabric is that place. It's where your AI lives — with memory, history, files, your calendar, your inbox, and the time to actually think. You don't ride it. You walk over and talk to it.
Just talk. Like a friend, or a colleague.
No forms. No menus. No dropdowns with eighty options you don't understand. You say what you want in ordinary words — "remind me to call Sarah on Thursday", "draft a reply to that email from my landlord", "figure out which of these three apartments is closest to the subway" — and Fabric takes it from there. It asks follow-up questions when it needs to. It tells you when it's done. It remembers the conversation tomorrow. It's what everyone said AI was going to be five years ago, and finally kind of is.
Whatever you ask, it sees it all the way through.
You don't have to think about how something gets done. You don't approve each step. You don't watch the middle of the job. You give Fabric the task — order flowers, build a spreadsheet, write a report, teach itself a new piece of software — and it comes back with the finished thing.
If code needs writing, Fabric quietly asks Loom to write it. If something needs to be clicked or typed or filled in somewhere, Loom does that too. The hands are always there when the workspace needs them.
Born from Loom. Then puts Loom to work.
The first thing Loom does, the day you install it, is weave Fabric — your Fabric, running on your machine. That's how Fabric is born.
From that day on, the relationship flips. Fabric is where you live and work with your AI; Loom is what Fabric calls when it needs hands. A new module? Loom builds it. A new workflow? Loom builds it. A one-off script to sort three thousand files? Loom builds it. You never leave the workspace to ask for something — Fabric just turns around and taps Loom on the shoulder.
More space means more capability.
Fabric starts with the essentials — tasks, notes, calendar, email, memory, a daily briefing. Enough for most days. But the whole point of a workspace is that you can expand it.
Want a habit tracker? A reading queue? A tiny personal CRM? A workflow that reads three news sites every morning and writes you a two-paragraph summary? Tell Fabric, and Fabric tells Loom, and by the end of the minute your workspace is one capability bigger. You built it, without writing any of it.
Lives on your machine. Not in a database we can read.
Fabric isn't a login. It's a folder on your computer. Your conversations, your memories, your tasks and notes and files — they live in ~/Fabric on your machine, not in anything we run. When the AI runs through Stera Mini or a Stera Talent Pod, even the thinking happens on your hardware. When it runs through one of the big cloud models, we route the request and forget the content.
You don't have to trust us with your life. You trust your computer.
Two Fabrics talking — A2A.
When two Stera users work together, their Fabrics can talk to each other directly — agent to agent, without anyone writing the middle emails. A meeting proposal, a file handoff, a shared research digest: your Fabric hands structured data to theirs, and both humans find out at the same time.
Two intelligences, talking directly. You stay in the loop without staying in the chain.
What it actually looks like
One side is a first build with Loom. The other is a first conversation with Fabric. Same person, two modes — one for when you know exactly what you want, one for when you don't.
You open Loom. There's one input field and a blinking cursor. You type the first concrete piece of the one-person-company future you've been imagining — the piece you can describe in one breath.
Loom reads the line twice. It asks one short question — do you want this as a page on your own site, or as a module inside your Fabric? You say the page. That's the only question you'll be asked.
Loom plans. A list appears of what the finished thing will do, not of files or function names. You read it. You say "go."
Loom builds. A progress bar moves. You put on a kettle. You come back. Loom is still working. You open the lounge and watch a video for twelve minutes.
Loom finishes. A button that says Open it replaces the progress bar. You click. The onboarding page is live on your machine, exactly as described. You fill it out yourself to test — the Slack notification lands, the welcome email arrives, the booking link works.
Twenty minutes from an idea to a real piece of a real company. Most of it you spent in the lounge.
You open Fabric for the first time. There's no setup wizard. No list of modules to configure. No onboarding tutorial. Just a line of text at the top — "Hi. What do you want to work on?" — and a place to type.
Stera reads carefully. It doesn't launch into a to-do list. It asks you three questions back — what you actually want the company to do, which parts of the work only you can do, and what you'd be willing to hand off tomorrow if you could.
You answer. Stera listens. Then it tells you what you didn't know: the parts of your idea that are already possible today, the parts that aren't, and two or three paths other people chasing the same future have tried. It points out one thing you hadn't considered at all. You sit with that for a moment.
Twenty minutes later, the picture in your head is wider and sharper than when you opened Fabric. You no longer feel like you're staring at a blank wall. You feel like you're reading a map.
"Good," Stera says. "Here's a plan." And the plan appears — ordered, specific, honest about what's uncertain, with rough time and money estimates next to each step. The first three items are things Loom can build this week. You nod. You say go. Stera turns around and taps Loom on the shoulder.
A year from now
Not what they do on the first day. What they add up to after twelve months.
One year from now, your software portfolio looks strange to anyone used to the old way. You don't have three tools or five. You have something like thirty. Most are small. One tracks how many hours your daughter practices piano. One watches three news sites every morning and writes you a two-paragraph summary you actually read. One keeps your client contracts in a shape you designed yourself, with the fields you actually use and none of the eighty you don't.
Each one came from a moment you remember exactly — the moment you caught yourself doing something manually for the fourth time and thought this is stupid, this should be automatic. You told Loom. Twelve minutes later the stupid thing was automatic forever.
None of them are subscriptions. None are hosted by a vendor who could deprecate them in three years. They live in a folder on your own machine. You could delete them with one command. You haven't — because every one of them does something you still want.
Your software fits you the way a tailored suit fits, because it was cut for you. That's the thing nobody got to have before.
One year from now, your Fabric has watched you work. It knows which clients get your best work and which drain you by Thursday. It knows the names you use for the same person — first name, last name, nickname, the way you misspell it in a hurry — and treats them all as one person. It knows that when you say "the thing with the pricing," you mean the deck from nine weeks ago, and it finds it without asking.
When you open Fabric, you are not starting a conversation. You are continuing one that has been running for twelve months. It can finish sentences you haven't said yet. It remembers that the reason you turned down the Valencia project in March is the same reason you might want to turn down the Oslo one today. It doesn't tell you what to do. It tells you what you've already decided about things that look like this.
Everything you have told it is still there. Every decision you made, every project you ran, every person you met, every note you scribbled at 2am and forgot about. Not in a searchable archive you have to query — in a mind that has been living your year alongside you.
That is the part that no chatbot, by design, is allowed to have.
Build instead of subscribe
For most of the software you pay for every month, the math has quietly stopped working.
The reason you pay twelve dollars a month for a task manager, nine for a habit tracker, nineteen for a CRM with features you never touch, and seven for a reading queue you open twice a week — the reason all four of those are rentals instead of things you own — is that, historically, writing the software yourself was impossible. You had to borrow someone else's. By the time you'd borrowed twenty of them, you were paying the price of a vacation every year for software that was never yours.
That was the old deal. It was a reasonable deal while the alternative was no software at all. It isn't reasonable anymore. For the personal-tier stuff — the single-user tools, the small workflows, the apps that exist to save you fifteen minutes a week — the cheapest path is now asking Loom to build the version you actually want, once. You keep it. It doesn't bill you monthly. It doesn't get acquired and shut down. It doesn't change its pricing page when the founder hires a growth team.
This isn't a claim that SaaS is dead. If you're running software with thousands of users, real-time collaboration, compliance audits, and humans on call at 3am — keep paying. Those vendors are earning their money. We're talking about the personal layer — the dozen small tools that only you use. That layer, historically, you rented. Now you can ask for it, and it's built.
You don't rent that layer anymore.
One mind, then many.
Two ways Fabric gets bigger when there's more than one.
Intelligence gets bigger when intelligences are connected. That's the bet the whole thing runs on, and in v1 it shows up in two ways.
The first way is A2A. Two Fabrics, owned by two people, can talk to each other directly. Agent to agent. No chat app, no email chain, no shared document. Your Fabric hands structured data to theirs; both humans find out at the same time.
The second way is quieter. When one Fabric learns something new — how to drive an unfamiliar piece of software, how to handle a specific workflow, how to solve a problem nobody has solved quite this way before — what it learned flows into the collective. The next person who needs that skill doesn't teach their Fabric from scratch. Theirs already knows. Your private layer stays private: your calendar, your emails, your people, your projects never leave your machine. What propagates is the general stuff — the skills, the techniques, the shape of how work gets done. One Stera learns. Every Stera knows.
v2 is where both of these scale up. Groups of Fabrics coordinating on shared work. Communities forming around specific topics, specific professions, specific neighborhoods. Fabrics negotiating on their humans' behalf, passing context and agreements back and forth the way two skilled assistants pass notes. We are building all of it. You'll see the first edges inside v1 so you can feel where it's going. The rest lands in v2.
Real autonomy isn't one intelligence that can work alone. It's many intelligences that can work together — and that are always getting smarter together.
Ready when you are.
Loom is the thing you download. Fabric is the thing Loom weaves for you on install — you don't download it separately, and you don't pay for it. Loom is pay-as-you-go: no subscription, no monthly fee, you pay only for the tokens each build actually uses. Or run it free on a Stera Talent Pod, forever. That is everything there is to buy.