brainplane and One, side by side.

Public beta download for macOS 13 or later, Apple silicon and Intel. Free, Pro and Lifetime plans are on sale. Newest build 0.3.19 on 8 October 2026. O1O2O5

Public download for macOS 26 or later on Apple silicon, signed and notarized by Apple. 14 days free with every feature, then $12 a month or $99 a year. B1B7B8

The home page lists Claude Code, Codex, OpenCode, Pi, Omp, Amp and Grok. The support page lists the same without Amp. O1O2

14 agents: Claude Code, Codex, Grok, OpenCode, Pi, Omp, Amp, Cursor CLI, Gemini CLI, Antigravity CLI, Copilot CLI, Qwen Code, Factory Droid and Aider. 12 of them take delegated work. Claude Code, Codex, Grok and OpenCode are tested live. B1B3B8

One shows agents from supported terminals. To answer, stop or message an agent from One, the session runs in herdr, a free open-source terminal for agents that setup installs. One can move an open terminal conversation into herdr when you send it a message. O1O2

Works through each vendor's own command line tool, its hooks and an MCP server. No extra terminal to install. B1B4B8

A public update feed with 117 builds. O5

Signed updates inside the app, checked once a day. An update or an update feed changed after signing is refused. B8

For agents in herdr, 1 key answers a numbered question from the Inbox. O1

1 key answers a Claude Code, Codex or Grok session waiting in iTerm2, tmux or herdr, after a 2 s hold you can cancel with Escape. Terminal tabs come to the front. Tested live with Claude Code in iTerm2. B1B4B8

1 list of questions and finished work from connected agents, answered with number keys. Finished work on the Mac can show up to 2 suggested replies. O1O4

1 card per topic, high risk first, including questions from live terminal sessions and suggested next steps. Routine questions wait in a batch with defaults you accept in 1 press. B1B2B8

Double-tap Option to speak or type. One routes the request to a matching session or launches an agent. It uses your ChatGPT plan or OpenAI key first, with One's AI as backup. O1

Speak or type from any app with Control Option B. brainplane places the request on the right topic and agent, offers the idle session that owns the topic, or starts a new run. Parsing runs on your own Claude Code plan. Nothing runs before you press. B1B5B8

A dot for every agent at the edge of the screen: working, needs you, done. O1

A count and a lamp per agent in the menu bar, an optional strip at the screen edge, and the Expert rail. B8

Number keys answer, J and K move between cards, E clears one. O1

Number keys and arrows in the Inbox. Expert mode adds panes, a command line and a go-to picker. B8

Finished work on the Mac comes with up to 2 replies, ready to send. O1

Finished work shows up to 3 next steps as buttons, written by the agent that did the work. brainplane makes no AI call for them. B8

Since build 0.3.18, opens without a microphone and lets you type. O5

The bar and the command line take typing. No microphone is needed. B8

Since build 0.3.17, moves Claude desktop app sessions so they can be answered. O5

Reopens a session with no terminal, such as a Claude desktop app session, in 1 press: same session, your reply sent as its first prompt, in iTerm2 or tmux. B8

Since build 0.3.19, trusts a new folder in 1 tap. O5

For Claude Code: shows the folder and its git remote, and 1 press answers. Outside your project roots it takes a second press. B8

To answer, stop or message an agent from One, the session runs in herdr. O1

Answer agents in iTerm2, tmux and herdr, where they already run. Terminal tabs come to the front in 1 press. B8

Not described. O1

Tiled panes for every agent, the Inbox, topic state and receipts on 1 screen, keyboard first, with a command line. B8

Not described. One keeps its own local history, work memory and summaries of agent threads. O2O3

1 shared state per topic that Claude Code, Codex and Grok read and write: where it stands, decisions, answered questions and files touched. Each handoff gets a context pack. B1B2B8

Not described. O1

Each handoff carries a goal, a context pack, a time budget and acceptance checks. A failed check goes back to the agent as a list of fixes, with a capped number of rounds. B1B2B4B8

Not described. O1

Hand a result to another model family for review in 1 action, for example a Claude draft to Codex, on your own plans. Tested live with Codex reviews on 8 and 9 October 2026. B1B2B8

Not described. O1

1 browser lease per profile with a queue and priorities, a parallel lane, and no takeover of your own browser. B8

Not described. O1

Every MCP server each agent sees, on 1 page. Add a server once for Claude Code, Codex and Grok. B8

Not described. O1

Your own agent reads Slack on a schedule and prepares a draft card. You send. B8

Not described. One keeps a local work memory. O2O3

When a question matches one you already answered, brainplane reuses your answer and shows its source on a receipt. Sensitive or close matches come to you for review. 1 rule set is compiled for Claude Code, Codex and Grok. B1B4B8

One acts after your request or your answer in the Inbox. You choose whether a suggested follow-up goes to an agent. O1

Each type of action starts at ask. After 3 matching approvals in 14 days, an action with a tested way to reverse it gets a promotion card, and you decide. 1 undo moves it back down. Sending, paying, deleting, deploying, account creation and paid plans always ask, fixed in code. B4B8

Each answer waits 2 seconds and Escape takes it back. One shows a receipt of what it did. O1

A receipt for every stored write. An answer waits 2 seconds, 5 for high risk, and you can cancel it in that time. Some topic edits undo for 10 seconds. Automatic answers can be marked undone for 24 hours, which also moves that action back to ask. An answer an agent already received stays received. B4B8

Not described. O1

Write a skill once. brainplane installs it for Claude Code, Codex and Grok in 1 press, with a backup and an undo. B1B6B8

Speak or type from any app with a double-tap of Option. Speech is transcribed by a cloud or Apple engine. You can drag over part of the screen, then speak. O1O3

Press Control Option B in any app and speak or type. Works with superwhisper, Aqua Voice and other dictation apps, Apple speech on the Mac, or OpenAI on your own key. Point at part of the screen and it goes with the request to your own agent only. B1B5B8

Not described. O1

Every session has a name, Alpha, Bravo or your own. Say "Bravo, review the letter" and it goes to that session after 2 seconds you can cancel. The name also shows on the tmux window and herdr pane. B5B8

Speech is transcribed by a cloud or Apple engine. O1O3

Parakeet v3, an open model by NVIDIA for 25 languages, runs on your Mac with the network blocked. Downloaded on your press. Apple speech, OpenAI and dictation apps stay available. B5B8

iPhone and iPad app as a free TestFlight beta, iOS 17 or later, with notifications. Other phones use a web page without notifications. O1

brainplane on your iPhone, installed from Safari to the Home Screen, with notifications for questions and finished work, end-to-end encrypted. Works on Android and any phone browser too. B1B8

Up to 8 Linux servers, x64 or arm64, with no open port on the server and end-to-end encryption to your devices. Agents there can get replies and files, and can be stopped or started. Since build 0.3.16 of 8 October 2026, 2 Macs on 1 account see each other's agents, and the 2nd can take over when the main Mac sleeps. O1O4O5

Add Linux servers, x64 and arm64, over SSH from your Mac with your own keys. No open port on the server, end-to-end encrypted. Agents there answer in tmux, with herdr support built in. B1B8

Presented for coding agents. Its public pages do not describe records for business topics. O1

Topics hold the state and decisions of contracts, email drafts, payment recaps, browser work and code. The work itself still runs through your agents. 1 agent at a time drives the browser. B1B2B8

Other phones use a web page without notifications. O1

The phone app runs in Android Chrome and desktop browsers, end-to-end encrypted, with Web Push notifications where the browser supports them. B8

Agents on a connected Linux server stay reachable from the phone when the Mac is off. Agents on the Mac need it awake. O1

When 1 Mac sleeps, your phone requests go to your awake Mac. A server's questions wait on the server and arrive when your Mac wakes. B8

Server agents must run in herdr. O1

Server agents run in tmux on Linux, with herdr support built in. Install over SSH with your own keys. B8

A request from the Mac can send your words, a screen capture and recent agent context to AI providers through One's service, or straight to OpenAI with your own key. Thread excerpts are also processed for titles and recaps. One says it keeps no server archive of that content. Linux server traffic is end-to-end encrypted to your devices. O1O3O4

The local brainplane service makes no AI call. Context goes only to the agent tools you run, on your own accounts, and each of them talks to its own vendor. When you turn on OpenAI speech, the Mac app sends audio to OpenAI with your key. The app checks its licence with brainplane.app, sending your licence, a device hash, the app version and the Mac name you choose. With a phone, a 2nd Mac or a server paired, messages pass through brainplane's relay end-to-end encrypted, so the relay cannot read them. Screenshots stay on your Mac and go only to the agent you run. B1B5B8

You sign in to a One account with Apple, Google or email. One's service handles AI requests and relays phone and server traffic. O3O5

1 brainplane account by email for your licence, Macs and billing. An optional relay for your phone, other Macs and servers, end-to-end encrypted. Your agents never need the account. B1B8

Usage events are linked to your account, and the app has no switch to turn them off. Prompts, audio, screenshots and terminal output are excluded. O3

None in the app: no analytics or crash reporting code. The website counts visits with Google Analytics only if you accept it in the cookie banner. B1B8

Sign in with Apple, Google or email. A plan paid on the website activates in the app. O1O3

Sign in with a 6-digit email code. My account shows your plan, licence key, Macs and invoices. B8

Its FAQ says only traffic to your Linux servers is end-to-end encrypted. O1

Every message between your phone, your Macs and your servers is end-to-end encrypted, Mac agents included. B8

Not described. O1

The lock screen only says that an agent needs you. The question shows once you open the app, decrypted on your phone. B8

A request sends a screenshot through One's server to AI providers, or to OpenAI with your own key. O1O3

A screenshot stays on your Mac and goes only to the agent you run. B8

Not described. O1O3

A password, a one-time code or an API key in a note is refused before it is stored, with a reminder to rotate it. IBANs are masked to their last 4 characters. The same guard runs before anything goes to your phone. B1B8

Free: 25 requests a day to One's AI. Pro: unlimited usage and premium routing, $12.99 a month ($6.99 the first month) or $119.88 a year, 3-day trial. Lifetime: $229 for the next 53 seats shown, then $299. On every plan, requests can use your ChatGPT plan or your own OpenAI key, with One's AI as backup. Your coding agents stay on your own plans. O1

$12 a month or $99 a year, 14 days free with every feature and no card. Founding price: $69 a year for the first 500 yearly buyers, kept while the plan renews. Your agents stay on your own plans. B7