COMING SOON · A DIGITAL ONE INITIATIVE
One conversation that wraps the coding agents you already use: Claude Code, Codex, Gemini CLI and more. Say it once; multiple agents pick up the work. Simpler than an IDE. Easy forever.
Illustration: a single message is typed once. A router guided by Gateward and FrontierScore signals splits the work — planning to Claude Code, building to Codex, research to Gemini CLI and private data to a local model — and the results rejoin as one result, ready for your review.
Your agents already run in different terminals, IDEs and browser tabs. builds.one brings them into one thread. You ask once; a planner breaks the job down, the right agents pick up the pieces, a reviewer checks the work, and the context travels with every handoff. You can step in at any point.
BRING YOUR OWN SUBSCRIPTIONS · WHAT WE'RE BUILDING
Illustration: five separate tool windows — claude code, codex, antigravity, ai studio and gemini cli — fold into one chat thread. A planner, a builder, a researcher and a reviewer pass context along four handoffs until a pull request is ready.
Name the tool and model you trust for a job, and every step goes there. Or switch on Auto: Gateward is designed to route each sub-task on live evidence from FrontierScore, such as quality, speed and cost, so no single agent has to do everything.
WHAT WE'RE BUILDING
All steps routed to Claude Code.
Every step goes to your pick.
ILLUSTRATIVEThe same conversation changes shape for the job. These are the modes we're building.
ROADMAP · WHAT WE'RE BUILDINGLong-run autopilot for real engineering work. Close the laptop; it keeps going in a DispatChai residence, a small always-warm EU Linux home for your agents, and picks up on your phone right where it was.
dispatchai.cloud The phone-native remote chat is being built there.
A quick question gets a quick answer, sent to whichever model is best for it right now. No setup, no tool choice.
Agents working next to you on documents, spreadsheets and tasks. A frontier model plans the workbook; a local model can fill it in on your own data.
Compose your own agent or assistant from routing rules, tools and models. Give it a name, a budget and a job, then share it with your team.
Two agents from different providers answer independently, then compare notes. Where they disagree, you see it.
Your policy decides what leaves the building.
Your organisation sets the rules: block certain questions, redact what shouldn't travel, and send sensitive work to your own local models. A frontier model such as Claude Opus or Claude Fable designs the strategy; an open-weights model such as Kimi, running on your hardware, reads the private data and follows the plan.
LIME = STAYED LOCAL
Illustration: a request to build a spreadsheet from client data passes through the organisation's policy. Client names and row data are redacted from anything that leaves; the strategy, with no data, goes to a frontier model, while the rows and instructions stay with a local model on your own hardware. The finished spreadsheet returns to the chat.
Give each person, team or the whole organisation a budget. Get warned before a run gets expensive, not after. One view of spend across every provider you bring, and an audit trail of which model saw what.
Example mock-ups: a team budget ring at 80 percent, an alert that a team reached 80 percent of its budget and a rule moved drafts to a lower-cost model, and an audit trail showing which model saw which data.
builds.one is the front door — the one conversation. Behind it sit the other pieces of the Digital One ecosystem, each doing one thing well, and one Digital One ID is how you’ll sign in.
Also on the list
We're building builds.one now. Early access opens by invitation through Digital One ID.
Request early accessInvite-only for now. No invite? You'll join the waitlist.