CLOSED BETA · A DIGITAL ONE PRODUCT
builds.one runs Claude Code, Codex, GitHub Copilot and its own builds coder where your code lives: your Mac, your Linux server or a hosted EU residence. The Orchestra makes them a team. Every screen you pick up is a live view of the same session. Easy forever.
By invitation: a request joins the waitlist, with your Digital One ID.
Illustration of how a turn travels: you type a message on any screen; the hub, run by Digital One in the EU or on your own server, passes it to a desk on one of your residences — a Mac running Claude Code or Codex, a Linux server on an API key, or a hosted EU residence running GitHub Copilot — and the result comes back to every screen.
A coding orchestrator that runs where the code is: on-prem on your own machines, or in the EU cloud. A residence is a machine that runs builds server. A desk is a project folder on it, with its own agent and its own conversation, streamed live. The agents work next to the code; your screens look in.
On-prem · on your machines
A self-hosted residence, installed with one line; on Windows it runs in WSL (Ubuntu). Your code, sessions, plans and keys stay on that machine, and it connects out over HTTPS only.
curl -fsSL https://app.builds.one/get | sh -s -- <code>
The app gives you the line with your own code: Residences → Add a residence.
In the EU cloud
A DispatChai residence in the EU, with nothing to install. It pairs with builds.one by itself and is driven from the same apps: DispatChai is where the agents live, builds.one is how you talk to them. Close the laptop; the work keeps going. A builds.one invite can come with one.
The hub that links your screens is run by Digital One: in the EU, or on your own server.
The session lives on the residence. Your phone, tablet, laptop and browser are live views of it, so you can walk away from any of them.
Sessions survive. Close the laptop or lose signal: the turn keeps going, and every screen catches up.
Push when it matters. Web Push and iOS notifications when an agent needs you or is done.
Approvals with the diff. Allow once, always, or reject, with the change in front of you.
Answer from the lock screen. On iPhone, with suggested quick replies for the next step.
Face ID first. Risky actions ask for it before they run.
Illustration: on a laptop, the desk "web" runs Claude Code. It fixes a flaky test, then asks to run git push, showing the diff with Allow once, Always and Reject. The same question arrives as a notification on an iPhone lock screen; it is allowed there with Face ID, the laptop shows "Allowed once, from iPhone", and a second notification says the turn is done, with suggested quick replies.
builds.one sells no model access. It drives the coding agents and keys you already have, and keeps an eye on their limits for you.
Coding agents · on your plan
API keys · our own builds coder
Keys are sealed on your device, for one machine, with spend caps per turn, per day and in total.
Several accounts per machine. builds.one rotates between your Claude and ChatGPT accounts, and resumes when a limit resets.
Plan-limit meters. See what is left on each plan, right in the app.
Switch mid-conversation. Change provider, model or effort; the context is handed over.
Past the limit, onto Copilot. When every Claude plan is spent, a desk can move to your Copilot plan first and, where your plan offers it, keep its model.
Not a wish list: all of this is in the beta today. The roadmap has its own section, further down.
Every desk keeps a to-do list you can edit from any screen. Switch on the auto-worker and it works through the list by itself.
Reports and write-ups an agent makes for you are kept with the conversation, announced with a push and badged until you open them.
Open the files a session works on, or send any file into the conversation.
Sub-agents and background agents show up in the conversation. Several desks can work in parallel.
Choose how much the agent may do on its own, and switch model or effort from one turn to the next. Slash commands go through to Claude Code.
Standing instructions and memory cards go with you from desk to desk. On Claude desks, “Ask the org” brings in SkilledMind.
Caps per turn, per day and in total for your API keys. The Usage page shows it all, Gateward and SkilledMind included.
Add an API key from any screen: it is sealed on your device, for one machine.
A FrontierScore card ranks the models you actually have.
Manage residences and desks from the app, with health checks and background services that update themselves.
The Orchestra is the virtual team manager on top of your desk. Helpers test, review and fix in rounds, your desk’s main session conducts them, and you watch every one of them live on the Score, from any screen.
NEW · IN THE CLOSED BETAHelpers with a role. More than 30 built in: builders, reviewers, testers, quality, ops and docs.
Rounds from playbooks. QA, fix, review and code-quality rounds come with it. Save your own, or keep them in the repo.
A conductor. Your desk’s main session runs the rounds, within what you grant it.
On a schedule, or on a trigger. Rounds start by themselves. Hold one helper, stop a round, and pick up again after a restart.
You, at the gate. Answer a helper from the podium, then approve its work or send it back.
Every helper works in its own worktree. Nothing is pushed or merged: you merge.
Helpers run in a sandbox, with the network off.
A machine cap, a time limit per helper, budgets and a kill switch.
On Claude, Codex and Copilot desks, on your own plans. Never on an API key.
The Orchestra is optional: a desk without it works as it always has, and only the machine’s owner starts a round. It is the first step toward VSETS.
Organisations use the Digital One suite’s roles and invites, on the same Digital One ID. Create one in the app, invite people by email, and give the team a residence.
An isolated seat for each member. On an org residence, members make personal desks, kept apart from other members’. Each runs on its member’s own plan.
An org Gateward key, capped. Optional, with a daily cap per member that the org’s admin sets.
SmartFrontier as a model. Gateward’s router is one more choice in the model picker.
FrontierScore evidence. A card ranks the models your team actually has, on FrontierScore’s live evidence.
“Ask the org”. SkilledMind brings what your organisation knows into a Claude desk: a short brief as each conversation opens, a cited answer card you can attach to your next message, and learning back when a session ends, with one switch per desk. Each member uses their own key, with a member’s rights; if SkilledMind can’t be reached, the session starts anyway and says why.
builds.one works on its own. Gateward, FrontierScore, SkilledMind and DispatChai are integrations, never requirements. Setting builds.one up for a team? info@digital1.one
The beta is closed, and tighter than before, on purpose. We visit each organisation, sign an NDA, and show what builds.one does with Gateward, SmartFrontier and SkilledMind switched on.
CLOSED BETA · UNDER NDACommunication encrypted end to end, across your teams and with your partner organisations.
Sessions your team can share, with the load balanced for you.
Highly redundant, with failover management and elastic scaling built in.
Secured, encrypted local sessions: passkey-locked, hidden, or set to disappear by themselves.
Controls kept in sync with your MSA and NDA.
YubiKey FIPS support, for the teams that need it.
Now open for pilots
VSETS
In those rooms, the conversation tends to move from “can it?” to “how much of our engineering could it take on?” VSETS is the next step from one desk at a time: dedicated teams of agent fleets, each team led by a human, at a flat monthly price, built around builds.one Enterprise Ultra. The Orchestra was the first step.
Compose a team at vsetaas.com. To see builds.one for your organisation: info@digital1.one
Shown under NDA, to organisations in the closed beta. The details stay in the room.
The apps are views; the residence does the work. Every platform is live in the closed beta. We are still testing, so the apps reach invited teams only: through TestFlight and our own test builds.
Any modern browser, installable
Live in the closed beta · app.builds.one
Push, lock-screen answers, Face ID
Live · TestFlight, by invitation
Your Mac can also be a residence
Live · by invitation (Apple silicon)
The builds.one app, and builds server to make your PC a residence
Live · test builds, by invitation
The same app as on iPhone and iPad
Live · test builds, by invitation
A residence for your server, run by builds server
Live · one-line install · tested on Ubuntu 24.04
WindowsLive · test builds
builds.one
Your conversations in a window of their own: a taskbar badge when a desk needs you, notifications while the window is hidden, and buildsone:// links that open a conversation. Closed, it waits in the tray.
builds server
Makes your PC a residence. The residence runs in WSL (Ubuntu) for now; the app shows its desks, lets you allow or reject a desk’s request from a notification, opens a terminal, and can start when you sign in.
AndroidLive · test builds
Built from the same code as the iPhone and iPad app: the same conversations, approval cards with the diff, files and documents. Push notifications on Android come next.
Also, in Chrome
Open app.builds.one and choose Install app: it opens like an app, with the same screens.
Live in the closed beta, and still in testing: invited teams get the Windows and Android builds from us, with no public download. app.builds.one also installs as an app from Edge or Chrome.
curl -fsSL https://app.builds.one/get | sh -s -- <code>
connects out over HTTPS onlycode, sessions and keys stay on the machine
What we are building next. None of it is in the beta yet.
ROADMAP · IN PROGRESSIn progress
Start from a template: one sentence with blanks to fill, the right practices offered with their reason, and a self-review that runs your repo’s own tools before the agent calls it done. Untick anything; only the blanks you leave empty become questions.
Next milestone
Low-level code from AI, behind a checker. An agent may emit LLVM IR, one assembly function, a Wasm module, an eBPF program or a GPU kernel only where a checker can judge it: proven, refuted, or asked. Every verdict keeps its evidence and a signed record, and your source stays the reference.
Roadmap
Two agents from different providers answer independently, then compare notes. Where they disagree, you see it.
Roadmap
Your organisation’s policy decides what may leave: sensitive work is rerouted to local models on your own hardware.
Roadmap
No local model to reroute to? Personal data, client names and your own IP are swapped for realistic look-alikes before a prompt leaves, then swapped back in the answer.
Roadmap
Agents working next to you on documents, spreadsheets and tasks.
Roadmap
Compose your own agents and assistants, then share them with your team.
Roadmap
Gemini CLI and OpenCode, through the Agent Client Protocol.
Roadmap
Desks that run on models living on your own hardware.
Roadmap
Built, and being redesigned before it ships.
Roadmap
VS Code, Zed and JetBrains as more views of the same session.
builds.one is the front door: one chat for your coding agents, on top of the Digital One suite. It works on its own, and the rest plugs in when you have it. One Digital One ID signs you in everywhere.
builds.one is in a closed beta, by invitation. Invited? Your invitation sets up your residence and first desk: open it on the web, or in the apps it brings.
Request access with your Digital One ID; you’ll join the waitlist.