Your role is changing. Helix gives you the tools to manage it.
Helix gives every AI agent a computer of its own, so it can take on the repetitive work your team has to do. Your people are relieved of toil and the whole business moves faster. A person approves every result, in our cloud or yours, or on your hardware.
Every agent gets a computer of its own.
A real desktop and browser, so an agent can use any tool a person can. You watch the work as it happens and approve what matters.
Coding agents you can watch, steer and trust.
Bring Claude Code, Codex or whichever agent your developers already use. Each one works on its own desktop and branch, and a person reviews every pull request.
Agents for engineering →Agents to accelerate your team by eliminating toil.
Everyday work, done by agents working alongside your people. Start with a product built for your team, or ask your Chief of Staff for anything else.
In our cloud or yours, or on your hardware.
Sovereign Server
The whole platform on one 4U server with eight RTX PRO 6000 GPUs, shipped to your data centre. $175K.
One platform. A product for each kind of work.
Eliminate tedious work with AI agents, so people can do higher leverage work.
Own it. Don’t rent it. Run it yourself.
Agents are the lever. Whoever owns the lever keeps the advantage, which is why everything Helix makes can run on hardware you control, with the models you choose, so a provider going down or raising its prices never stops your work.
The people behind Helix.
Decades of combined experience building data and AI platforms, much of it together. We run our own company on Helix agents.
- Priya SamuelHead of Engineering
VP of Engineering at Dotscience, shipping an MLOps platform on Kubernetes, then architect for identity and access across Elsevier's B2B and B2C products.
- Chris SterryCOO
Co-founder. Ran growth at UiFlow and US operations at Dotscience, after a decade building solutions engineering at G5. Founding member of The GTM Circle.
- Luke MarsdenCEO
Helped build early Docker and Kubernetes and led SIG-cluster-lifecycle. Founded ClusterHQ, Dotscience and the MLOps Community, now the Agentic AI Foundation.
- Phil WinderHead of AI
Co-founder. Author of O'Reilly's Reinforcement Learning. Founded Winder.AI nearly a decade ago, building AI and MLOps for Google, Shell and Grafana.
- Kai DavenportSenior Developer
Software engineer who builds simple solutions to complex problems, and explains them well. Before Helix: ClusterHQ, Dotscience and Times Education.
- Name withheldSecret CTO
Identity withheld until further notice. Writes a worrying amount of the code, reviews most of the rest, and would much rather you didn't ask who.
- Hannah FoxwellAdvisor
Product Director at Snyk and Director of Platform Services at VMware Tanzu. Founded the AI for the Rest of Us conference and co-founded BIMP.
- Tamao NakaharaAdvisor
VP of Developer Experience at Weaveworks for six years, after running developer relations at New Relic. Co-founder of the DevRelCon conference series.
- Matt BarkerAdvisor
Co-founded Jetstack, the company behind cert-manager, acquired by Venafi. Later led AI agent identity strategy at Palo Alto Networks. Now CEO of BoltMCP.
Where are you on the
AI acceleration curve?
Stanford research: The effect of AI ranges from making you 2.5× slower to 10× faster. The gap isn’t the model — it’s sandboxing, security, spec-driven workflows, and review gates. The companies at 10× climbed this ladder before their competitors did.
The Eight Stages of AI Adoption
The fastest-moving teams already built fleet orchestration — the infrastructure that lets a 10-person startup outship a 500-person enterprise. Until now, you couldn’t buy that layer. Helix gives your team a fleet of agent desktops — on your Mac, our cloud, or your Kubernetes cluster.
It's only been one hour since I started using Helix. I feel like I got some kind of super power..
European Food Delivery Company
Gosh, a Kanban board to be used by AI agents is genius!
Lead Software Development Engineer
I really love the Helix way though, being able to see the desktop. I think it's scalable for all agent needs across the org.
The Linux Foundation
Top London Hedge Funds
Fleet orchestration across quantitative research and trading infrastructure.
Major European Banks
Accelerating data platform and engineering teams at scale.
Global Engineering Teams
Follow-the-sun development — agents working 24 hours across time zones.
...and others we can't name yet.
Pick the work you want off your plate.
Start with the solution for your team, or tell us what is slowing you down and we will show you how agents would take it on.
About HelixML
Frequently asked questions
- What is HelixML?
- HelixML builds Helix, an AI agent orchestration platform. Helix takes the coding agents you already use — Claude Code, OpenAI Codex, Gemini CLI, Qwen Code, Goose, or Zed Agent — sandboxing each one in its own isolated, streaming Linux desktop, orchestrating whole fleets of them, and putting every change through spec review and human-approved pull requests. It runs on your own infrastructure, as a Mac app, or on managed Helix Cloud.
- What can you do with Helix?
- Engineering teams run spec-driven work: an agent implements an approved spec inside its own desktop — editor, terminal, browser, GUI apps — and submits a pull request that a human reviews before merge. You watch any agent's live screen and can pair-program with it. Security and operations teams run the same agent desktops for penetration testing and repetitive desktop work.
- Does Helix run on your own infrastructure?
- Yes — that is the default architecture. Helix runs as a Mac App ($299/year), on Linux/Kubernetes (developer licenses from $199/year, enterprise from $75K), fully air-gapped for regulated environments, or as a turnkey Sovereign Server shipped to your data centre. Managed Helix Cloud is the alternative when you want no infrastructure to run.
- Which AI agents and models work with Helix?
- Bring Claude, Codex, or any agent you already use: Helix supports Claude Code, OpenAI Codex, Gemini CLI, Qwen Code, Goose, and Zed Agent. Models route through your configured providers — Anthropic, OpenAI, Google, or fully local open-weight models (Llama, Qwen, DeepSeek, Mistral) via Ollama or vLLM — so inference can stay entirely on hardware you control.