Offrun offers one Mac workspace for running multiple coding agent CLIs

What Offrun does

A new macOS application called Offrun, featured on Hacker News, wants to be the single place where developers run the coding agent CLIs they already have. According to the project's site, the app does not replace your agents or supply models of its own. Instead, it runs existing CLIs directly on your Mac with each one signed in as you, so there is no new account to create, no API key to hand over, and no password visible to the app.

Getting started is a three-step flow: install the app, connect your agents through a settings pane (adding a second account if you have one), then pick a project and give an agent a task.

Isolation through git worktrees

The central technical idea is that every agent working in a project gets its own git worktree. Because each worktree is a separate checkout, two agents operating in the same repository never write to the same files. The site's example is running a refactor and a bug fix simultaneously, then merging each result on your own schedule.

A second agent reviews the diff

Offrun also builds peer review into the workflow. You designate one of your agents as a reviewer, and it reads the diff while the work is still uncommitted, posting its findings into the chat. Suggested fixes arrive in your message box so you can read them before anything is sent, and nothing merges without your approval. According to the site, this runs on your own agent accounts, meaning no second subscription and no automated bot leaving comments on your pull requests.

Surviving rate limits

For users juggling multiple subscriptions, Offrun lets you point each agent at a different login. When one login hits its usage limit, the app moves the conversation to a login that still has capacity and tells you it did so. Resend your message and the new login continues with the full conversation history; if every login is exhausted, Offrun offers a different agent instead.

Memory, notifications and the wider workspace

Several supporting features round out the app:

• Project memory: repository-wide conventions live in one place, and each agent's goal, plan and dead ends in another. Agents read this context when a session starts and write back to it as they work, so a fourth session knows what the first one already tried. Everything is stored as plain files in your project folder, which you can open, edit and commit like any other file.
• Notifications: the app watches coding agents anywhere on your Mac, including ones running in terminals it never launched, and alerts you when an agent finishes or stops to ask a question.
• Preview: changed files, diffs with line numbers, rendered markdown and tables, images, PDFs and a small built-in browser appear beside the conversation that produced them.
• Modes: a Build mode has the agent make the change and run it to prove it works, while a Planner mode works out an approach without touching a file. Any chat can switch between the two at any time.
• Dictation: hold a key and speak your prompt. Transcription runs entirely on-device, and the site says nothing is recorded or uploaded.

Privacy, price and requirements

According to the site, dictation and your project files stay on your Mac, and agent prompts travel straight to whichever provider you are signed into, under your own account. The developer states that Offrun has no server in that path and never handles your keys, and that the network is only needed for initial setup. Everything is free with no feature limits, no seat counting and no credit card. Requirements are an Apple Silicon Mac (M1 or newer) running macOS 14 Sonoma or later, plus at least one agent CLI you already use.

Why it matters

Developers increasingly run several agent CLIs side by side, and the friction shows up in predictable places: two agents editing the same checkout, wasted effort because session four forgets what session one learned, and conversations that stall when a subscription limit kicks in. Offrun targets all three directly — worktrees for isolation, plain-file memory for continuity, and login failover for limits. Its free, local, bring-your-own-account stance also contrasts with hosted orchestration layers that sit between you and the model provider, which may appeal to teams wary of another service handling their code. The caveats are worth stating plainly: every claim here comes from the product's own launch page, so the usual Show HN scrutiny over whether it works as advertised will follow, and the Apple Silicon-only requirement leaves Windows and Linux developers out for now.