zcoder

Parallel-first AI coding harness

A parallel-first AI coding harness built in Rust, focused on orchestration, workbench-oriented workflows for production coding.

Get Started Star on GitHub

Why zcoder?

Most coding harnesses are useful, but they were not designed to maximize AI and human parallelism. zcoder is built parallel-first from the ground up: workbench-centric, worktree-native, and focused on real work thread isolation. A workbench is the unit of work. By supporting multiple workbenches per worktree, zcoder keeps files and context separated, runs coding tasks concurrently, and lets you review changes before they are integrated.

Parallel Execution Maximize throughput by running multiple coding tasks concurrently.
Workbench Oriented Key work unit per worktree, built for full work isolation and parallelism.
Native Worktrees Git worktrees out of the box. Isolation without the overhead.
Control First Inspect everything. Configure everything. Trust, but verify.
Performance Native Built in Rust for speed, reliability, and low overhead.

Getting Started

# Built it Rust
cargo install zcoder

# run it
zc

# Right now, just starter code, nothing much 0.0.1

zcoder origins

zcoder originates from the experience of pro@coder, an AIPack created by Jeremy Chone. It brings the control-first, magic-second and workbench-centric philosophy of pro@coder into a dedicated TUI and production-grade, parallel-first coding harness. Core ideas like context globs, knowledge globs, and workbenches evolve into stronger concepts for isolated parallel execution, native worktrees, and reusable coding workflows.

Why Z ? because I still cannot pronounce th.