I am Josh and I always bounce between Windows, Mac, and Linux, and I use coding agents from different providers. I couldn’t find a tool I actually liked that worked the same way on all three, or even had a distribution for all three for that matter, so I decided to build Yardsort and open-source it.
It’s a desktop app: every task gets its own git worktree and a real terminal running the agent CLI you already have (Claude Code, Codex, Cursor, OpenCode, Grok, …). And you can start several agents in parallel sharing the same worktree or branch while working on the same task.
After the first version worked, I started using Yardsort to build Yardsort. That’s how most of the “nice” features showed up. They were just what my own workflow needed. I basically followed the pain.
A lot of tools let you swap agents on the same worktree or branch. Fine until you need the next agent to know what the last one actually did. Context sharing is usually a pain. So I built handoff: it writes the next agent’s first message from the task, commits, changed files, and what was recorded about the runs. It is not a dump of the previous chat. There’s also project memory you approve before any agent sees it.
I was also doing the same chores over and over. For example, “have another agent review what this one just shipped.” So I built the workflows feature. Workflows are YAML you write yourself, or you can ask an LLM to draft one and keep what you like. Built-in example: request a code review from a second agent in the same worktree.
Everything is private by default: no account, no telemetry, no keys of mine. I’m currently working on “Yardsort sync” so you can share memory and do handoffs across machines via a cloud service with end-to-end encryption, and I’m open-sourcing that path too, with an easy self-host option, because I don’t want this to only work my way.
1 comment
[ 4.6 ms ] story [ 31.6 ms ] threadI am Josh and I always bounce between Windows, Mac, and Linux, and I use coding agents from different providers. I couldn’t find a tool I actually liked that worked the same way on all three, or even had a distribution for all three for that matter, so I decided to build Yardsort and open-source it.
It’s a desktop app: every task gets its own git worktree and a real terminal running the agent CLI you already have (Claude Code, Codex, Cursor, OpenCode, Grok, …). And you can start several agents in parallel sharing the same worktree or branch while working on the same task.
After the first version worked, I started using Yardsort to build Yardsort. That’s how most of the “nice” features showed up. They were just what my own workflow needed. I basically followed the pain.
A lot of tools let you swap agents on the same worktree or branch. Fine until you need the next agent to know what the last one actually did. Context sharing is usually a pain. So I built handoff: it writes the next agent’s first message from the task, commits, changed files, and what was recorded about the runs. It is not a dump of the previous chat. There’s also project memory you approve before any agent sees it.
I was also doing the same chores over and over. For example, “have another agent review what this one just shipped.” So I built the workflows feature. Workflows are YAML you write yourself, or you can ask an LLM to draft one and keep what you like. Built-in example: request a code review from a second agent in the same worktree.
Everything is private by default: no account, no telemetry, no keys of mine. I’m currently working on “Yardsort sync” so you can share memory and do handoffs across machines via a cloud service with end-to-end encryption, and I’m open-sourcing that path too, with an easy self-host option, because I don’t want this to only work my way.
GPL-3.0, Tauri + Rust. Early, moving fast, rough edges expected.
Site: https://yardsort.sh
Repo: https://github.com/joaoh82/yardsort
Downloads for all three OSes via GitHub Releases.
Happy to dig into handoff, workflows, or any other bits if you have questions.
Cheers, Josh