There is no single download.
Products in this platform are delivered differently. One installs into an editor, one runs in infrastructure you control, one runs without installing anything at all, and one is not distributable yet.
This page names each path, says plainly which of them work today, and ends every one of them somewhere real.
Five paths, one platform.
The delivery model follows from what the thing is. A runtime that has to be in the path of an operation installs where the operation happens. An assessment that reads evidence needs nothing installed at all.
- Assessment
- Nothing is installed. It runs against a deployment and returns a result, including the parts it could not answer. Anyone evaluating, at any depth. The shortest path to real output.
- Editor install
- One command registers the runtime with an editor as an MCP server. Operations then pass through it. Developers and platform teams putting governance in the path of a coding agent.
- Self-hosted service
- You run the service in your own environment, from source or with Docker Compose. There is no hosted tier to sign up for. Organisations and public-sector institutions standardising across teams.
- Packaged installer
- A downloadable file. None exists, for any product in the platform, and there is no release date for one. Nobody yet. Named here because its absence is what people arrive looking for.
- Personal application
- A desktop or mobile application on the usual stores. Not built, not submitted, not listed. Individuals governing their own AI assistants.
Two editors, one command each.
These are the surfaces this repository ships install scripts for and tests against. The runtime is editor-agnostic and speaks MCP; an editor appears here only once its install path exists and has been run.
Claude Code
python scripts/install_claude_mcp.py
Registers the runtime with the editor. Operations then pass through it before they execute.
Cursor
python scripts/install_cursor_mcp.py
The same runtime, the same records. Which editor was used is part of the record rather than a difference in what is recorded.
The runtime is a server and the editors are clients. Claude Code and Cursor are the two this repository installs for and tests against, not the two it depends on. Any MCP client can be a client; those two are the ones with a script, which is the difference between AVAILABLE and a statement about the protocol.
- Command line
- The
vaicommand line runs an audit and writes a receipt to disk. It arrives with the service rather than separately: see Install, then the quick start. - Self-hosted service
- Python from source, or
docker compose up -d --build. Both paths are documented end to end, including the health check that says it worked. - Browser extension
- Records what an assistant did in the browser, on your machine. Loaded from source. There is no store listing for it, and none is claimed. See Install Sentinel.
There is nothing on this page to download.
No packaged installer exists for any product in the platform, and there is no release date for one. The rows below are what is being built for; each names the thing it is waiting on.
- Runtime, packaged A signed installer for the runtime
- Waiting on packaging and signing. Until it exists, the paths above are the whole of how the runtime is obtained.
- VerifAIer Me · Desktop macOS, Windows, Linux
- Waiting on the same packaging work. Nothing can be downloaded and there is no release date.
- VerifAIer Me · Apple App Store iPhone, iPad
- Waiting on the desktop surface first, then review. A local-first product on a phone is a different design, not a smaller window.
- VerifAIer Me · Google Play Android
- Waiting on the same mobile design work.
Rows rather than buttons, and no dates anywhere. A button that cannot be pressed says the page is broken; a row says the work is not finished. When a surface is ready its row becomes an action, in the same place, without the page moving.
Every path ends somewhere real.
Four readers arrive at this page. None of them should leave it without one thing to do.
Run an assessment
Nothing to install and nobody to contact first. It returns real output, including the answers it cannot give.
Install into your editor
One command, then operations pass through the runtime. The two commands above are the whole of it.
Talk about a deployment
Deployment shape, adopted policy packs and evidence residency are decisions about your environment, not a form to fill in.
Sovereign deployment
The same architecture, inside your own perimeter. There is no separate government product to obtain.
The shortest honest path is the one that installs nothing.
An assessment runs against a deployment you control and returns evidence you keep, including the parts it cannot answer. Whatever else eventually becomes downloadable, that will still be the fastest way to find out whether any of this is true.