Helix is Tray’s platform for building production apps with AI coding agents. You describe what you want, an agent such as Claude Code writes real TypeScript, and helix deploy puts the app live on Tray’s cloud. IT teams govern every project from one dashboard: who can open each app, which authentications it uses, and a trail of what changed.
Building this way, describing an app in plain language while an agent writes and runs the code, is sometimes called vibe coding. On Helix the result is ordinary code in an ordinary repo, running on infrastructure Tray manages.
Who it’s for
Three groups work with Helix, often on the same projects.
Business builders. You drive the agent and may never hand-write code. Describe the app, review what it does, and deploy it. Hosting and credentials are already handled.
Developers. A Helix project is standard TypeScript and React. Open it in VS Code, use git and npm, and modify agent-written code like any other code. Project structure shows the layout.
IT and security teams. Every project, deployment, and access change is recorded and visible in the dashboard, and execution logs show the third-party calls each request made. You control who can open each app and which authentications a project can reach. The security model covers this in depth.
What’s included
Every project gets the same managed pieces, with no setup work:
| Included | What you get |
|---|---|
| CLI and MCP server | @trayai/helix-cli plus the MCP server it runs, which Claude connects to so the agent drives the project through the same commands you do |
| Key-value store | Project-scoped storage that persists between runs on a deployed project |
| Authenticated third-party access | Function code names a service by alias; the credential is resolved server-side and never appears in your code |
| One-command deploy | helix deploy builds the app and ships it to a managed runtime |
| Builder dashboard | Your projects, deployments, access settings, and execution logs at app.helix.tray.ai |
| Admin console | The org-wide view for org admins: projects, deployments, authentications, and recent activity |
Apps are TypeScript. The app type Helix clearly supports today is a single-page application; API-only patterns are still being confirmed. Every deployed app runs behind login: a visitor has to be a user in your Tray or Helix organization who has been granted access to that project, so nothing you deploy is public. Deployed apps live at https://{project-id}.helix-app.ai.
Locally, helix dev runs the project on your machine. Local key-value data is held in memory and resets when the dev server restarts; only a deployed, provisioned project keeps it.
Why real code
LLMs have seen TypeScript, React, and HTTP APIs millions of times in training data; proprietary visual builders, they haven’t. Helix keeps its conventions minimal so agents stay reliable: the file path is the URL path, the method suffix is the HTTP verb, and input validation is a Zod schema next to the function. When the agent is done, a developer can read and change everything it wrote. How Helix works explains the architecture behind this.
Availability
Helix is pre-GA. Design Partner slots are open now for teams with a real deploy need who want to shape the product, and Early Access opens in the second half of August 2026. Both start with a conversation: talk to your Tray contact, or request access at helix.tray.ai.
Helix is currently switched on in a limited set of Tray orgs: Main Tray Org, Early Access Org, SE Demo Org, and Workspace test - Platform. If your org isn’t one of those, your Tray contact arranges access.
Features marked with a roadmap callout, or listed as shipping soon on the limits and roadmap page, aren’t available yet. Everything else in these docs works today.
Next step
Work through the quickstart: install the CLI, log in, and deploy your first app.