yyork

Run a fleet of AI coding agents in parallel — each in its own durable, isolated workspace.

---

yyork spawns AI coding agents into isolated, durable workspaces and lets you supervise them from a single dashboard. Point it at a git repo, hand an agent a prompt, and it goes to work in its own worktree while you start the next one.

• Parallel — every session gets its own git worktree and branch, so agents never step on each other.
• Durable — sessions outlive your browser, a server restart, and even the agent process exiting; Zellij keeps the pane alive so you can always read what happened.
• Local-first — it all runs on your machine. State is one SQLite file at ~/.yyork/state.db — no account, no cloud.
• Bring your own agent — drives the Claude Code and Codex CLIs, unchanged.

Requirements

• Go 1.25+ and Node.js 22+ / pnpm — to build the server, CLI, and dashboard
• Zellij — hosts the durable sessions
• git — sessions run in per-session worktrees
• An agent CLI on your PATH — Claude Code (default) or Codex
• Nix with flakes — optional, for the dev shell

Install

yyork ships as a single binary that embeds the dashboard — build it once:

git clone https://github.com/yyopc/yyork.git && cd yyork
pnpm install
pnpm backend:build      # build + embed the dashboard, compile ./yyork

Drop ./yyork somewhere on your PATH, or run it in place.

Quick start

# 1. Start the dashboard + API on 127.0.0.1:7331 (opens your browser)
./yyork

# 2. From any git repo, hand an agent a task
cd ~/Projects/my-app
yyork spawn --prompt "add a health-check endpoint"

# 3. Watch it work — click the session in the dashboard, or attach from a shell
zellij attach <sessionId>

# 4. When you're done
yyork session list          # what's running
yyork stop <sessionId>      # kill the session, remove its worktree + branch

spawn creates a yyork/<sessionId> worktree off your repo's default branch, launches the agent in a fresh Zellij session, and the session appears in the dashboard within milliseconds (live over SSE — no polling). From the dashboard you can open a live terminal, rename a session, jump into your IDE, and stop it.

Warning

There's no in-app review or merge yet, and both yyork stop and a reboot delete the yyork/<sessionId> branch. Push or merge before you stop — e.g. have the agent run git push -u origin yyork/<sessionId> or gh pr create.

Agents

yyork drives real agent CLIs; choose one per session with --agent:

• claude-code (default) — Claude Code
• codex — Codex

Other spawn flags:

• --prompt <text> — the task for the agent (required)
• --permissions <mode> — default | accept-edits | auto | bypass-permissions
• --system-prompt-file <path> — a system prompt to launch the agent with

FAQ

How is this different from running Claude Code or Codex directly? It's the layer around them. Those CLIs run one agent in your working tree; yyork runs many at once — each in an isolated worktree on its own branch — keeps them alive in Zellij, and gives you one dashboard to supervise the fleet. The agents themselves are unchanged.

Where does the work end up? On the session's yyork/<sessionId> branch in your repo. Push or merge it before stopping the session (see the warning above).

Does anything leave my machine? No. The server binds to 127.0.0.1 and all state is a SQLite file under ~/.yyork. Your agent CLI talks to whatever model provider you've configured it with — yyork doesn't sit in the middle.

Development

For dashboard work, run the dev stack instead of the binary — Vite serves with hot reload and proxies /api to the Go server:

nix develop && pnpm install && yyork   # Vite on :3000 + Go API
# or, without Nix:
pnpm dev

Run pnpm test for the backend, CLI, and frontend tests, and pnpm run to list every script (including the e2e:live-terminal suite).

---

MIT © yyopc