Manage every coding agent from one workspace.

offrun.dev · 2026-10-04 · by Long Nguyen ·generic
Run Claude Code, Codex and Antigravity in parallel on your Mac, each in its own git worktree. Pick a second agent to review the diff. Free.

Manage every coding agent from one workspace.

Get started

Bring the coding harness that you already use.

Offrun runs the CLIs on your Mac, signed in as you. No new account, no API key, and it never sees your password.

1. Install Offrun. Drag it to Applications. 2. Connect your coding harness account. Settings, then Agents. Add a second account if you have one. 3. Start an agent. Pick a project and give it a task.

Agents

Each agent runs in its own folder. Offrun never sees your password.

Claude CodeAdd account

Defaultsam@studio.dev

Worksam@acme.co

CodexAdd account

DefaultSigned in with ChatGPT

AGYAdd account

DefaultSigned in with Google

Grok BuildAdd account

DefaultSigned in with X

Pi and OpenCode are coming soon.

Side by sidePeer reviewLimits

01 · Parallel agents, one git worktree each

Agents that never step on each other.

Every agent in a project gets its own git worktree, so two agents working the same repo never touch the same files. Run a refactor and a bug fix at once and merge them on your own schedule.

Projects

billing-svc

api-gateway

offrun-site

---

Agents

Fix flaky upload test_now_

Retry backoff_1m_

Refactor the parser_2m_

Token refresh_18m_

billing-svc fix-flaky-upload

Fix flaky upload testRetry backoffRefactor the parserToken refresh

The upload test fails about one run in ten. Find out why and fix it.

The test checked the file before the upload had finished, so it raced the network. It now waits for the upload itself instead of a fixed timer.

Changed `test/upload.test.js` only, on this chat's own branch. It passed 50 runs in a row.

Ask to make changes

✳︎ Opus 5.5 (1M)Build

02 · Code review by a second agent

A second agent reads the diff before you do.

Pick a reviewer, and its findings land in the chat while the work is still uncommitted. Ask for the fixes and they land in your message box, for you to read before anything is sent. Nothing merges on its own.

Runs on your own agent accounts. No second subscription, no bot on your pull requests.

Projects

billing-svc

api-gateway

offrun-site

---

Agents

Fix flaky upload test_3m_

Retry backoff_now_

Refactor the parser_4m_

Token refresh_20m_

billing-svc api-retries

Fix flaky upload testRetry backoffRefactor the parserToken refresh

Cap the retry backoff at 30 seconds and add jitter.

Done. Retries now wait at most 30 seconds, plus up to 20% jitter so they spread out.

js

async function retry(req, n) { const wait = Math.min(2 * n 100, 30_000) const jitter = wait 0.2 Math.random() await sleep(wait + jitter)}

Review by Codex · changes requested

1. Warning in `src/retry.js:2` The 30 second cap is longer than the gateway’s 25 second timeout, so the last retry is cut off.

Ask the agent to fix these

03 · Multiple accounts and usage limits

Hit a limit, keep working.

Point each agent at a different login. When one hits its limit, Offrun moves the chat to a login that still has room and says so. Send your message again and the new login picks up the whole conversation. Out of logins, it offers another agent.

Projects

billing-svc

api-gateway

offrun-site

---

Agents

Fix flaky upload test_6m_

Retry backoff_2m_

Refactor the parser_now_

Token refresh_23m_

billing-svc parser-split

Fix flaky upload testRetry backoffRefactor the parserToken refresh

Split the upload parser into a reader and a validator.

This login has hit its usage limit.

Keep going with another agent

Claude Code resets 3am. Codex can pick up this chat. It gets the whole conversation.

Continue with Codex

Ask to make changes

✳︎ Opus 5.5 (1M)Build

The workspace

What the workspace gives you.

Project memory

Your agents stop forgetting

Offrun keeps repo-wide conventions in one place and each agent's own goal, plan, and dead ends in another. Agents read that context at the start and write back to it as they go, so the fourth session knows what the first one already tried.

Stored as plain files in your project folder. Open them, edit them, commit them.

Memory · api-gateway

`conventions.md`Every agent reads it

`plan.md`This feature's approach

`dead-ends.md`Tried, failed, recorded

Read at the start of every session · written back as they work

Preview

Read the work without leaving the chat

Changed files, diffs with line numbers, rendered markdown and tables, images, PDFs, and a small browser with an address bar, all beside the conversation that produced them.

Fixed. The uploader reported done as soon as the first upload finished, so the second was sometimes still running.

#### What changed

  • `src/uploader.js` now waits for every upload.
  • The test checks which files arrived, not their order.
  • 2 rows

    | Run | Passed | Failed | | -------------- | -------- | ------ | | Before the fix | 41 of 50 | 9 | | After the fix | 50 of 50 | 0 |

    ---

    Dictation

    Talk to your agents

    Hold a key and speak your prompt. Offrun transcribes on-device, keeps your identifiers spelled the way your code spells them, and drops clean text wherever your cursor is.

    Transcription runs entirely on your Mac. Nothing recorded, nothing uploaded.

    Mac book MacBookVoice ready, hears “mac book”

    jason JSONNo voice samples

    Add voice sample

    retry backoff helper retryBackoffHelper1 of 2 voice samples

    KubernetesVocabularyNo voice samples

    Add voice sample

    Privacy and security

    Your Mac is the boundary.

    1. ### Your voice stays here. Dictation turns speech into text on this Mac. No recording is kept or sent. 2. ### So do your projects. Chats, memory and worktrees are folders in your repo, under `.offrun` and `.worktrees`. We keep no copy. 3. ### Prompts go straight to your provider. Each agent signs in as you, so your code goes from this Mac to Anthropic, OpenAI, Google or xAI. Offrun has no server in that path and never handles your keys.

    Read the privacy policy

    This Mac

    DictationSpeech to text, on device

    Never leaves

    billing-svc

    Chats, memory, worktrees`.offrun` · `.worktrees`

    In your repo

    Claude

    Prompts and codeSent on your own login

    Leaves, direct

    On your login

    Your providerAnthropic · OpenAI · Google · xAI

    Offrun serversNot in this path

    Pricing

    Free. All of it.

    Bring your own agent subscriptions. Everything Offrun adds around them is free.

    Free

    Everything, unlimited

    $0

    Every feature, no limits, no seats to count, no card.

  • Unlimited agents and projects
  • Unlimited on-device dictation
  • Unlimited connected coding agent accounts
  • Free credits for 100+ connectors
  • Your agents can find an email, file a GitHub issue, or post to Slack. You choose which chats use each account.
  • Unlimited agent monitoring and reviews
Download for Mac

FAQ

Questions

Does Offrun replace Claude Code or Codex?

No. It runs them. You keep your own accounts and subscriptions, and Offrun gives them somewhere to live together.

Can I use more than one Claude Code account?

Yes. Add each login in Settings, then Agents. When one hits its usage limit, the chat moves to the next and says so, and you send your message again. How multiple accounts work

Can Claude Code and Codex work on the same repository?

Yes, at the same time. Each agent gets its own git worktree and branch, so they never edit the same files, and one can review the other's diff. Using them together

Does it work with Google's Antigravity CLI?

Yes. Offrun runs `agy`, the Antigravity CLI, signed in with your Google account, next to Claude Code and Codex. Running agents in parallel

How is Offrun different from Conductor?

Both run agents in git worktrees on your Mac, on your own subscriptions. Conductor builds in pull requests, checks and team features. Offrun is built around several logins per agent for usage limits, a reviewer you pick from your own agent accounts, and Antigravity next to Claude Code and Codex, and it is free. The full comparison

Do you see my code?

No. Your code goes from your Mac to whichever provider you are signed into, under your own account. Offrun has no server in that path and never handles your keys.

What does peer review actually run on?

One of your own agent accounts. Pick the model in the model picker.

Can peer review change my code?

No. It reads the diff and posts what it found. Every decision after that is yours.

Where does project memory live?

In your project folder, as files. Version them if you want.

Does dictation need a network connection?

Only to set up. After that it runs on your Mac.

What do I need to run this?

An Apple Silicon Mac running macOS 14 Sonoma or later, and at least one agent CLI you already use.

Run them all. Miss nothing.

Free and unlimited. No credit card.

Download for Mac

Apple Silicon only (M1 or newer) · macOS 14 Sonoma or later

offrun.dev · 5 images · 14 links

Comments (0)

No comments yet