Z EngineA coding agent with a desktop app

Describe a task and Z Engine reads files, runs commands and edits code, asking for your approval before gated operations.

Get started

How it works

Download

One permission gate

Every tool, hook and check goes through one permission gate. Approve gated actions one by one, or write rules that approve them for you.

Multi-agent orchestration

Built-in and custom subagents run in parallel, nested, in the background or resumed, each with its own model, tools and permission mode. An agent can work in its own git worktree and hand back changes you apply or discard.

Claude Code compatible

Tools, commands, hooks and .claude/ folders follow Claude Code, so existing agents, commands, skills and CLAUDE.md files work as-is.

Steering and interrupts

Messages you send while the agent works are injected at the next step; nothing is dropped while an approval waits.

Evidence-backed badges

Every turn that changes files is marked Verified, Unverified or Failed from the checks that actually ran.

Durable checkpoints

Code, including shell-made changes, and the conversation can be rewound to any prompt; sessions survive crashes.

Models, caching and cost

Native Anthropic caching and extended thinking, OpenAI-compatible providers, fallback models, and per-agent cost tracking.

Optional sandbox

Confines shell commands to the workspace.