VVADRA

LOCAL-FIRST · DEVELOPER PREVIEW

Start setting up VADRA in about a minute.

One control plane for AI work across your machines.

LocalGPU runnerCloud runner

Challenge / repair → Independent re-verify → Proof gate

Set up VADRATry free demo

Start with the setup steps; installation and environment checks can take longer.

Use VADRA with your own project.

Existing Developer Preview: Node.js 20.9+, pnpm 11.9, Python 3.11, uv and Git. Protected execution requires macOS sandbox containment; the Ubuntu public site supports only the safe sample and replay.

Already have a source installation? Continue below. The signed installer for new users is not released and source downloads are retired; the free browser sample is available now.

  1. 1. Launch VADRA locally (Developer Preview)Run `pnpm dev:all` to start today's local build — the signed installer is Planned, not available yet.
  2. 2. Complete local onboardingThe local app walks you through setup on this machine — there is no VADRA account or sign-in today.
  3. 3. Choose your projectThe Runner can only see workspaces you explicitly approve.
  4. 4. Connect your AI toolsOfficial Codex / Claude Code sign-in stays in the official tools.

Open VADRA

Handbook →

VADRA LOCAL RUNNER · PREVIEW / EARLY ACCESS

The signed Runner installer is not available yet.

VADRA is moving to this account-plus-Runner installation. There is no sign-up form yet; availability will appear on the Roadmap first, and this page will link the real installer and early-access sign-up when they exist. Nothing here downloads or pretends to install software today.

Join early access — watch the RoadmapLaunch channel (YouTube)

Developer Preview: run it locally today

If you already have the VADRA source on this machine, run `pnpm dev:all` to launch it locally in Developer Preview — "Start with my project" will detect that running local instance.

Signed macOS / Windows installer: Planned, not available yet.

What stays on your machine

Your project files stay local by default. The Runner receives access only to workspaces you approve, executes bounded commands inside them, and returns structured results and evidence — it never silently uploads your repository.

Your provider logins remain managed by the official Codex / Claude Code tools on your machine. VADRA never asks for your Claude or OpenAI password, cookie, token or API key.

Content is sent to an AI provider only for tasks you start, with the provider you chose, under that provider’s documented behavior. Cloud execution is a separate capability you would enable explicitly — it is never a silent default.

Why there is no source download any more

Earlier previews offered a full developer source archive. That distribution has been retired: the commercial architecture keeps VADRA’s proprietary orchestration and verification policy server-side, and ships customers a signed Runner instead of the application source tree. The retired archive is preserved privately for audit and its old URLs now answer 410 Gone.

Existing preview installations keep working locally; VADRA never remotely deletes or damages anything on your machine.

How the Runner, cloud and your workspace relate →

Current limits

The live artifact-review customer journey is still blocked in acceptance. Plan review and successful implementation tests do not certify an artifact. Required missing proof must still refuse protected action. The account, entitlement and cloud control-plane services described in the architecture are planned — a design document does not make them available.

Read the scoped roadmap →

Prefer to explore without installation? Open the free browser sample