Skip to content

Zwischen Docs

Inline adversarial review for AI coding agents. Zwischen attaches a second AI model to your coding session and watches what the agent does in real time – tool calls, file writes, stop attempts. The reviewer can block dangerous operations before they execute, reject file changes before they hit disk, or force the agent to keep working when it tries to stop too early.

  • Getting Started

    Install Zwischen, wire it into your coding agent, and run your first monitored session.

    Installation

  • How It Works

    Understand the reviewer's event stream, enforcement decisions, and how blocks get resolved.

    Mental model

  • Configuration

    Control which tools the reviewer sees, set up per-project rules, and configure profiles for different agents.

    Settings file

  • In a Session

    Steer the reviewer while you work – switch its profile or model, clear blocks, and watch its verdicts live.

    Session control

What is Zwischen?

Zwischen is a real-time adversarial reviewer for AI coding agents. It installs as a set of hooks in your coding agent and streams every tool call, file write, and agent decision to a second AI model for live evaluation. The reviewer – called Frank – issues enforcement decisions (block, release, or pass) that the client enforces locally before the agent can proceed.

Two hosts are supported: Claude Code and Codex. Your settings, profiles, and reviewers are shared between them.

The name comes from zwischenzug – the chess "in-between move" that disrupts the expected line.

This is not post-hoc review. The reviewer is watching the session unfold and can intervene while work is still in progress.

What you need Where to go
Install and configure Installation
Run your first session First session
Set up Codex Codex
Understand the review model Mental model
Configure reviewer behavior Settings file
Watch the reviewer live Live monitor
Something not working? Common issues