# VOID, the reversible autonomy layer for AI agents

> VOID is the reversible autonomy layer for AI agents. It classifies what an
> agent is about to do, holds the calls that cannot be undone, puts a person
> in front of them, and records every decision in a hash-chained ledger.
> Version 1.0.1. The TUI and the MCP server are shipped. The desktop app is in
> beta for macOS, Windows and iOS, by application. Every figure on this site
> is illustrative, and none of it is a measurement of a running system.

## What it claims to do

An agent makes a call. VOID decides the class of that call before it happens:
can it be undone exactly, can it only be mitigated, or is it irreversible. The
first two run. The third is held and shown to a person, who approves or refuses.
What was decided, by whom and when is written into a hash-chained ledger.

The pitch in one line: give an agent the authority to act, and give it the
records to be held to.

## What is real and what is not

A product that exists and a product whose every number is real are two
different claims, and this page makes both.

- **Real**: the web site, its documentation, and the waitlist behind the beta
  request form. Request counts for this deployment are real.
- **TUI**: Built and shipped. The terminal interface, with VOID inside it as
  the reverse tool it started as.
- **MCP**: Built and shipped. The MCP server, with the VOID tool ready to
  register with the agents that call tools.
- **Desktop app**: In beta. There is no download: you apply, and an invitation
  arrives. The desktop app is for macOS, Windows and iOS. There is no Linux
  build, and there is no promise of one.
- **Every figure is illustrative.** The class ladder, the ledger chain, the
  interception loop, the delivery census, the risk spread, the compensation
  trace and the terminal preview are drawn from example data. The console is a
  fixture, not a recording.

## How to read this site

- [/docs](https://www.voidagent.tech/docs): the documentation. Each page also has a markdown
  form at the same path with `.md` appended, which is what an agent should fetch.
- [/spec](https://www.voidagent.tech/spec): the specification.
- [/glossary](https://www.voidagent.tech/glossary): every term the site uses, one sentence each,
  at [/glossary.md](https://www.voidagent.tech/glossary.md).
- [/openapi.yaml](https://www.voidagent.tech/openapi.yaml): every path, operation and schema of
  the HTTP API this site serves, as OpenAPI 3.1.
- [/openapi.json](https://www.voidagent.tech/openapi.json): every path, operation and schema of
  the HTTP API this site serves, as JSON.
- [/when-to-use.md](https://www.voidagent.tech/when-to-use.md): the questions people ask before
  they adopt an agent guardrail, and the boundary of each answer.
- [/registry](https://www.voidagent.tech/registry): the call registry, what is classified and how.
- [/security](https://www.voidagent.tech/security), [/compliance](https://www.voidagent.tech/compliance): the
  guarantees and their limits.
- [/changelog](https://www.voidagent.tech/changelog): what has changed.
- [/cs](https://www.voidagent.tech/cs): the same site in Czech.

## Where to ask

- hello@void.systems: general
- security@void.systems: security reports
- press@void.systems: press

This summary is written for agents and is not a transcription of the page, which
is a marketing page whose content is interleaved with motion. The documents
below are generated from the same data the pages render, so they cannot disagree
with them.