builds.one · How it compares
Your agents. Your machines. One session you can see.
Run the coding agents your team already uses on machines you choose. See and approve their work from any device, and cap what API keys can spend.
We checked this page on 5 November 2026 against each product's public documents. We did not run the other products. This field moves fast, so follow the links. Something out of date? Tell us at info@digital1.one and we fix it.
The short version
Keep the agents you chose
Claude Code, Codex and GitHub Copilot in one place, plus builds coder, our own agent, on your API keys. Switch provider mid-conversation and the context comes along. We sell no model access.
Run them where you decide
Your Mac, your Linux server (tested on Ubuntu 24.04) or a hosted EU residence. Sessions keep running there, and credentials stay on that machine.
Approve the work from anywhere
A question from an agent becomes a card with the diff: allow once, always, or reject. Phone, tablet, laptop and browser all show the same session.
Know what it costs
Caps per day and in total on API keys, and per turn on builds coder. Plan use shows as tokens. Money shows as money.
What sets us apart is the combination.
A session record you hold, across agents from different vendors, on the connections you already have, with a company that stands behind it. Every single part has a strong rival. We did not find another commercial product that documents all of it as a generally available feature. Coder comes nearest: its self-hosted Agents keep chats in your own database and, as an experimental feature, delegate to Codex or Claude Code. Factory is also near: its agent loop runs on your machine with your own models, but it is one vendor's agent.
What this means exactly: your conversations stay on your machine, and our server relays them without writing them to disk. Cloud models still receive your prompts. This is about custody and control. It is not secrecy.
Next to the open-source tools
This is a strong field, and we respect it. All of these tools are open source, and they have public communities. builds.one is not open source. It is a product in closed beta from a company you can hold to account. What we offer is a clear session contract and the people behind it.
| Tool | What it is | Why teams like it | Where builds.one differs |
|---|---|---|---|
| Orca | Desktop app and server | Runs agents on your own machines. iOS and Android apps in beta, and a very large community. MIT. | Every agent's questions arrive as allow-or-reject cards with the diff. Orca's Chat UI shows question cards too, and it can launch agents in Yolo or Manual mode. Sessions resume after a machine restart. Orca keeps agents running across an app restart, but a host reboot ends them. API keys have spend caps. |
| Paperclip | A company of agents | Budgets at org, agent and project level. Org charts, ten named agent integrations and custom adapters. MIT. | A different job. Paperclip manages agents like staff. builds.one is the session your engineers work in. |
| Ruflo | A harness inside your agent | Multi-agent coordination inside Claude Code or Codex, with memory across sessions. MIT. | A different layer. Ruflo lives inside an agent. builds.one holds the session around the agent, on every device. |
| Brigade | A coordination ledger (escoffier-labs) | Local files, no daemon, with handoffs, receipts and gates. MIT. | A different layer. Brigade tracks tasks, claims and evidence across tools, and can run managed work under budgets and approval gates. builds.one runs the live session and its approvals. The older brigadecore/brigade is Kubernetes scripting, not an agent tool. |
| Paseo | Local daemon | Seven agents plus ACP, and native phone apps. Apache-2.0. | We did not find spend caps in its documents. Paseo Hub has organisations with seats and invitations. builds.one adds caps, isolated seats and hosted EU residences. |
| T3 Code | Server and background service | Seven agents in its README, plus Kiro, Pi, Muse Code and ACP agents in its docs; iOS and Android apps, approval cards and checkpoint rollback. Fast growth. MIT. | Very close in shape. builds.one adds organisations with roles, isolated seats, hosted EU residences and a vendor who answers. |
| Happy | Daemon and phone apps | Own harness plus Claude Code, Codex and any ACP agent. A colleague can join a session. MIT. | Spend caps, and organisations with roles and isolated seats. Happy has team servers with individual sign-in and one owner. We did not find spend limits in its documents. |
| OpenCode | Own agent loop and server | Many model providers, a permission API, revert and fork. A very large community. MIT. | builds.one drives the vendors' own agents as well as our own. Our phone and tablet apps come from us. |
"We did not find" means our search of the public documents found nothing. The feature can still exist.
Next to the commercial platforms
These platforms are ahead on certificates, identity integration and managed cloud. Most can also run tools on your machine, so that alone no longer sets anyone apart. The question to ask is where the agent loop and the session record live, and which vendors' agents you can use.
| Product | Where tools run | Where the loop and session live | Models | Choose it when |
|---|---|---|---|---|
| builds.one | Your Mac or Linux server, or a hosted EU residence | Your machine; our server relays and does not write conversations to disk | Claude Code, Codex, GitHub Copilot; your API keys on builds coder | You run more than one agent, and you want one governed session across them |
| Cursor | Cursor cloud or your machine | Cursor cloud (documented for Cloud Agents) | Grok, plus third-party models or your own keys | You want a first-class editor, worker pools and a full set of certificates. |
| Claude Code | Local, Anthropic VM, or your runner (beta) | Local in local mode; through Anthropic for remote | Claude | You use Claude only and want first-party tools and managed VMs. |
| Codex | Local, SSH host or OpenAI cloud | Local with a relay; OpenAI for cloud | OpenAI by default; also custom providers and local models | You work mainly with OpenAI and want its OS-enforced sandbox. |
| GitHub Copilot | GitHub sandbox, runner or your device | GitHub for cloud sessions; Copilot CLI can run on your machine with your own model | Several vendors chosen by GitHub; your own keys or local models in Copilot CLI; Claude and Codex agents in public preview | Your work already lives on GitHub and you want its reach and residency options. |
| Devin | Cognition cloud or dedicated VPC, or your machine (Devin CLI; Local Hands rolling out) | Cognition for cloud sessions; Devin CLI runs locally | Several vendors | You want autonomous cloud sessions and SOC 2 and ISO 27001. |
| Factory | Laptop, CI, VM or VPC | Loop on the machine that runs Droid. Control plane at Factory, except in airgapped mode; session storage follows your settings in Factory Private | Several vendors, your keys, self-hosted | You need a documented airgapped deployment from a vendor with SOC 2 Type II, ISO 27001 and ISO 42001. |
| Amp | Local or orb; self-hosted orbs | Amp, US-based | Several vendors, your keys | You want no seat fee on Teams and no markup on models outside Enterprise. |
| Coder Agents | Your own infrastructure | Your own infrastructure | Your choice | You want a model-agnostic, self-hosted platform and already run Coder. Generally available since Coder 2.37 (September 2026). |
What you get from a company our size
One family of products
One Digital One sign-in. Add hosted residences, key routing, model ranking or org knowledge from our own products when you want them. They stay optional.
The people who build it
We are a small team and we release often. In a pilot, you speak directly with the people who build builds.one, and your requests reach them first hand.
Business control as a founding rule
Credentials, approvals and spend limits have been design rules since the first commit. Organisations, roles, isolated seats and an org audit list build on them.
Who it fits today
- Software houses, agencies and ISVs with roughly 20 to 300 engineers, EU first: many client repositories, mixed tools and rules on how code is handled.
- Any team running two or more agents whose AI coding spend has started to surprise them.
- Engineering and AI-adoption consultancies that want to stay neutral between model vendors.
When another product is the better choice
We prefer to say this early. Look at Factory, Coder, GitHub Copilot or Cursor first if you need any of these today, and check each vendor's documents, because no one of them offers every item: SSO with your own identity provider, a SOC 2 or ISO 27001 report, SIEM export, an air-gapped deployment, a native Windows server, Jira or Linear triggers, or vendor-managed cloud sandboxes with no machine to run.
What a pilot looks like
Six weeks, with one or two teams of 5 to 15 engineers on 2 or 3 non-sensitive repositories with working CI. builds.one is free today, up to 20 seats per organisation. You pay your model vendors directly. We do not promise a speed multiple. A pilot proves control first, with your numbers.
| Week | Work | Result |
|---|---|---|
| 0 | Agree scope, data handling and exit criteria. Name a decision maker and a cost owner. | Written scope |
| 1–2 | Set up. Record a baseline from git history. Set caps. | Baseline; restart and approval drills pass |
| 3–5 | Live use. We log your requests and what we ship for them. | Weekly evidence |
| 6 | Decide together: extend, expand, stop, or pause to fix. | Written result, with its limits stated |
What we measure: risky actions that went through an approval, spend per merged change, change failure rate, lead and review time against the baseline, and work lost at a disconnect or a restart. The target for unapproved actions and for lost work is zero.
What we do not claim
We hold no ISO 27001 or SOC 2 certificate, and no independent penetration test has been done so far. builds.one is in closed beta and has no service level yet. Our security page gives the full list. We would rather you read it now than find it later.
See it on your own repositories.The closed beta is by invitation. Tell us about your team and we will shape a pilot with you.