Straight answers.

brainplane is a Mac app with a small local service. It gives your AI agents 1 shared brain per topic: Claude Code, Codex and Grok, and 11 more agent tools. It hands work from 1 agent to another with a contract and checks. It answers questions you already settled. What still needs you arrives in 1 Inbox, as 1 card per topic. It runs on your Mac. It never sends, pays, deletes or deploys on its own.

People who run AI agents every day, on code and on business work: contracts, email drafts, payment recaps, hiring, browser tasks.

It helps most when you run 2 or 3 agents and spend time copying context between them. It also helps with 1 agent that keeps asking the same things.

Yes. Download it from https://brainplane.app/download/, open it and start 14 days free with your email. No card.

A topic is 1 piece of work you follow: a deal, a trip, a repo, a hire, a dispute.

Each topic keeps its state in 4 blocks (done, waiting on you, waiting on others, next step), plus decisions, answered questions, people and the files agents touched.

If you already keep Markdown notes per project, brainplane can import them read-only. It changes 0 bytes in your files outside its own marked block, and only on topics where you turn that on.

No. brainplane sells no AI model, no AI subscription and no AI credits. It does not resell, pool or share anyone's access.

You sign in to each agent with your own account, through the vendor's own login. The brainplane price covers brainplane only.

At least 1 AI agent that runs on your own account:

• Claude Code, with a Claude plan or Anthropic API access
• Codex CLI, with a ChatGPT plan or an OpenAI API key
• Grok Build CLI, with your xAI account
• Plans, prices and usage limits are set by each vendor. Check them on the vendor's site. Work that agents do through brainplane counts against your own plan, like any other session.

Yes. With 1 agent you get topics, context packs, standing answers, receipts, the Inbox and the skills library.

Each added model family adds more. With 2 or 3 you also get handoffs between them and reviews by another family, for example Codex checking a draft Claude wrote. 1 subscription is enough to start. More subscriptions let brainplane spread work and reviews across them.

No. It never reads Claude, Codex or Grok login files, tokens or browser cookies, and it never asks for an agent API key.

Sign-in opens the vendor's own login in Terminal. brainplane then runs the vendor's status command and keeps whether you are signed in, plus a few non-secret fields such as the login method. It keeps no email or account name.

The 1 key brainplane can hold is your own OpenAI key for voice, if you choose that option. It stays in the macOS Keychain and is read only when you record.

The local service makes no AI call. Agents run only when you start a run or press Run on a card.

The exceptions are yours to turn on:

• Voice through OpenAI: audio goes to OpenAI on your own key, from the Mac app, with a daily cap of 30 minutes.
• Request sorting: when the command bar cannot place a request on its own, 1 short call to your own Claude Code or Codex sorts it. You can turn it off in Settings.

brainplane runs each vendor's official command line tool, unchanged, on your Mac and your own login. It uses the same commands you could type yourself. It never shares a plan between people. It runs no agent on brainplane's servers: a Linux server you add with bp node runs the vendor's tool there, on your account.

Each vendor's terms apply to your account, so read the terms of your plan.

No. brainplane starts the official, unchanged command line tools: claude -p, codex exec and grok -p. It passes the context pack on standard input and reads the result. Your agents keep their own updates, settings and logins.

The agent decides, with its own login settings. brainplane adds no API key to any run: a run gets the environment of the brainplane service plus 3 brainplane variables (topic, handoff and socket).

If you have both a plan and an API key set up for 1 agent, check that vendor's docs to see which one it uses.

brainplane trains no model. Its only server for your work is the relay for a paired phone, Mac or server, and it carries sealed messages it cannot read. What your agents send to their vendors follows each vendor's data and training settings for your account. Check those settings for your plan.

Every run is visible and bounded. A handoff carries a time budget and a capped number of review rounds, and each context pack holds 3,000 tokens at most.

Runs count toward your plan like any other session. When an agent reports a quota or login error, brainplane shows it on the card.

4 ways:

• 1 brain per topic: agents read the topic's state and decisions, so you stop briefing each one.
• Standing answers: a question you already settled gets your saved answer, with a receipt.
• Batching, when you turn it on: routine questions from 1 run arrive as 1 card per topic, with defaults you can accept in 1 press.
• Checked handoffs: work that fails its checks goes back to the agent with a list of fixes before it reaches you.

brainplane cuts the tokens agents spend on briefing and on going back and forth:

• Context packs: each agent starts a topic with a short brief of 3,000 tokens at most. Binding items come first; older notes are trimmed first.
• Delta packs: after the first pack, an agent gets the changes since its last one, plus the rules and settled decisions it must always see.
• Standing answers: a repeat question is answered from your saved answer, with no new round of work.
• Checks before results reach you: a failed check sends a short fix list back to the agent.
• brainplane does not publish a token saving figure. bp stats shows what it did for you.

A short brief for 1 agent on 1 topic: the goal, where things stand, settled decisions, answered questions, people and files. It holds 3,000 tokens at most, counted at about 4 characters per token.

You choose which topics may go to which vendor. A topic with no scope stays out of every pack. You can preview a pack before it goes.

Packs go to the agent on its standard input, never on the command line. brainplane keeps a copy for 30 days by default (you can set 1 to 365), then deletes it.

A decision you already made, saved with its source. When an agent asks the exact same question and it is not sensitive, brainplane replies with your saved answer and logs a receipt.

Sensitive questions, close matches and similar wording come to you as a proposal. You confirm before anything is reused.

Work that moves from 1 agent to another under a contract: a goal, a context pack, a budget and acceptance checks.

The checks run before the result reaches you. A failed check sends a fail list back to the agent that did the work. A reviewer from another model family can also check the result. Both share 1 capped budget.

The result lands as 1 finished card with the verdict, the checks, where the result is and the command to resume the session.

Each kind of action has a level: ask you, run with a receipt, or run quietly. Everything starts at ask.

After you approve the same kind of action 3 times in 14 days with no undo, and only if that action has a tested way to reverse it, you get 1 card proposing a step up. You decide. 1 undo moves it back down 1 level.

Send, pay, delete, deploy, account creation, paid plans and secrets stay at ask forever. That rule is fixed in the code.

The bp stats command reads your local log and shows: questions asked per day, the share answered without you, batched questions, stop cards, drafts, and how many delegations went through brainplane.

Never on its own. Sending, paying, deleting, deploying, creating accounts, buying paid plans and handling secrets always stop at a card for you. No setting moves them to automatic.

An agent can prepare the email draft, fill the form up to the card field, or write the payment recap. You press the final button yourself.

Yes. brainplane's own rule covers what brainplane does: it never sends, pays, deletes or deploys. What an agent can do depends on the agent's permissions.

In runs brainplane starts, those permissions follow the mode. Review and research runs are read-only: the read-only sandbox for Codex, a read and search tool list for Claude Code and Grok. Write runs let the agent edit files.

Agents you run yourself keep the permissions you give them. A stop card asks an agent to wait for you; the agent's own settings decide what it can do.

Yes. Topics hold contracts, email drafts, payment recaps, hiring, admin and browser work as well as repos. The agents do the work; brainplane keeps the state, the decisions and the receipts.

Drafts are tracked from creation until you mark them sent. When a task needs the browser, 1 agent holds it at a time and the others wait in a queue.

brainplane records which session touched which file. A session can claim a task or a file for 30 minutes, and brainplane flags it when another live session writes the same file in that time. For the browser, 1 agent holds it at a time and the others wait.

brainplane does not lock files on disk. Agents you start yourself can still write anywhere their own settings allow.

Every write brainplane stores leaves a receipt: what changed, who asked, and when.

The full log is plain text on your Mac, so you can audit every write.

Each answer you give waits 2 seconds, or 5 seconds when the risk is high, so you can cancel it. Some topic edits can be undone for 10 seconds. An answer that ran on automatic can be marked undone for 24 hours, which also moves that rule back down. An answer an agent already received cannot be recalled.

• A Mac with Apple silicon (M series). Intel Macs are not supported.
• macOS 26 or later.
• Nothing else to install for brainplane itself: the app carries its own Node runtime.
• At least 1 AI agent on your own account. Setup can install the missing ones for you.

Download the disk image from https://brainplane.app/download/. It is signed with an Apple Developer ID and notarized by Apple. Drag the app to Applications and open it.

The first run has 3 screens. On a Mac where your agents are already installed and signed in, that is 3 presses of Return. Then enter your email to start the 14-day trial, or paste your licence key.

• Finds the agent tools on your Mac (14 known) and checks whether each is signed in, with the vendor's own command.
• Offers to install a missing agent with the vendor's official installer, in the background, on your press. No admin password, the source is checked, and you get a receipt.
• Opens the vendor's own login in Terminal when an agent is not signed in.
• Shows the exact entry it would add to each agent's settings, the backup path and the undo command. Nothing changes until you press Start and connect.

The uninstaller is a script that undoes each install step from its own log, newest first. It removes the app, the background service, the bp command and the agent entries, and puts your earlier settings back from the backups.

Your brainplane data folder stays until you delete it yourself.

• Microphone and speech recognition, only if you use the built-in voice engines. With a dictation app, none.
• The Documents folder, only if you set up the optional superwhisper mode.
• Automation for iTerm2 and Terminal, asked once, so brainplane can type the answer you pressed into the iTerm2 session you picked, or bring the Terminal tab to the front.
• Screen Recording, asked once, the first time you point at part of the screen.

No. The code is proprietary. Every change to your agent settings, your skills folders or an install is shown first and keeps a backup. Every write brainplane stores leaves a receipt in a local log you can read.

Yes for brainplane itself: topics, the Inbox, receipts and settings all run on your Mac. Your agents need their vendor's service to work, as they do today. After activation, your licence works 30 days without a connection.

14 agents: Claude Code, Codex, Grok, OpenCode, Pi, Omp, Amp, Cursor CLI, Gemini CLI, Antigravity CLI, Copilot CLI, Qwen Code, Factory Droid and Aider. Claude Code, Codex and Grok are tested live on real accounts.

Any tool that speaks MCP can read and update the shared brain through the brainplane MCP server.

Yes. 1 key answers a live Claude Code, Codex or Grok session in iTerm2, tmux or herdr, with the 2 s hold and Escape. A Terminal tab comes to the front for you to answer. Before it types, brainplane checks that the session is still the one you picked, that it is not at a shell prompt and that the same question is still on screen. Codex sessions also take typed messages through Codex's own queue.

Yes. brainplane needs no special terminal. It counts the agent sessions running in your terminals, in the background, in desktop apps and in herdr, and works alongside the tools you already use.

A skill is a written procedure your agents can follow, such as reviewing a contract or filing invoices. brainplane lists, read-only, the skills it finds in the agents' skill folders and in folders you add: your own, public libraries, third-party and the ones agents ship with.

For each skill it shows which agent can see it, which can run it, and any problem, such as a description cut short or a duplicate name.

Yes. brainplane reads them where they are and never edits the originals.

Make it work everywhere installs 1 copy in a shared folder that Claude Code, Codex and Grok all read. Before anything changes you see the diff. The current version is copied to a backup folder first, and you can undo for 24 hours.

brainplane can follow a public skills library on GitHub. It compares what you have installed with the library and updates on your press, from a pinned and reviewed version. It shows every changed file first.

Yes. The New skill helper asks you what the skill should do and which past session it should learn from. That session's own agent writes the draft; brainplane never reads your session history itself. brainplane checks the draft, then a dry run shows what the skill would do.

brainplane can also run a skill's steps itself, with a gate before each step that needs you and a receipt for each step.

No. Any step that sends, pays, submits a form, invites someone or creates an account belongs to you. brainplane shows 1 card that says what to press and where, then waits. A skill that gives such a step to an agent is refused when brainplane builds it.

Only what you approve, each time with the exact change shown first and a backup kept next to the file:

• 1 MCP server entry, named brainplane, added with each agent's own command.
• Optional hook entries: in Claude Code settings and in Codex hooks. Grok reads the Claude Code hooks, so it needs no separate entry.
• Your other MCP servers, hooks and settings stay exactly as they were.

It gives each agent 10 tools: find the right topic, read its brief, update its state, log a decision, ask a question, raise a stop, register a draft, post a card, report a handoff as done, and link its session to a topic.

Agents can open cards. Only you can answer them.

A hook is a short shell script your agent runs at key moments: session start, your prompt, a tool use, the end of a turn. It passes that event to brainplane on your Mac.

Each hook has a time limit of 1 or 2 seconds and fails open: if brainplane is not running, the agent carries on as normal. It adds a little time per event; on the maker's Mac a Claude Code hook took about 90 milliseconds.

Codex asks you once to trust new hooks before it runs them.

• The agent connection can be undone from the app for 24 hours.
• Hooks: bp hooks uninstall claude --yes and bp hooks uninstall codex --yes remove only the brainplane entries.
• Or copy the backup back. Each change says where its backup is.
• The uninstaller undoes everything at once.

Yes. Open the command bar with Cmd+K in the app, or with Control Option B from any app. Type or dictate a request.

Every request becomes 1 card first, with a suggested agent and topic. Nothing runs until you press Run.

The paths:

• Your dictation app. brainplane recognises 7 sources: superwhisper, Aqua Voice, MacWhisper, Wispr Flow, VoiceInk, Spokenly and Apple Dictation. With a dictation app, brainplane needs no microphone access.
• OpenAI speech on your own OpenAI key, kept in the macOS Keychain. OpenAI lists it at about $0.0045 a minute (pricing page read on 2026-10-08). Daily cap: 30 minutes.
• Apple speech on the device: free, offline, once its language files are installed.
• Parakeet v3 on the device: free and offline, the default when it is ready.
• With no dictation app, Settings links to 3 of them on their official pages. brainplane installs no dictation app.

Aqua Voice and MacWhisper keep their shortcut in a file that also holds a login token. brainplane never opens files like that, so it cannot read the shortcut.

Record your shortcut once in Settings, Voice. brainplane then uses it in its hints. Your Aqua Voice shortcut keeps working in the bar as usual.

Only if you choose OpenAI speech: then the Mac app sends the audio to OpenAI with your key. Apple speech on the device keeps it on your Mac. A dictation app follows its own privacy policy.

brainplane never speaks back and never keeps audio.

On your Mac, in brainplane's own private folder in ~/Library/Application Support. Every change is 1 line in an append-only log. A local index is rebuilt from that log when needed. Your topics are not stored in a brainplane cloud. Sealed cards for a paired phone pass through the relay, which cannot read them. Your brainplane account holds only your email, your licence, your Macs and your invoices.

The brainplane service listens on your Mac only (127.0.0.1 and a private socket) and opens no connection to the internet itself. Traffic leaves your Mac in these cases:

• Agent runs. The agents you run, including the ones brainplane starts, talk to their own vendors under your account: your prompt, the context pack, the files they read and their tool calls. Each vendor's terms apply.
• Audio to OpenAI, only if you turn on OpenAI speech.
• On your press: a vendor installer download, a vendor sign-in, or a check of a public skills library on GitHub.
• A licence check from the app, at launch and once a day when online. It sends your licence, a device hash, the app version and the Mac name you choose, and no topic, file, prompt or agent data.
• Messages to your paired phone, Macs or servers, through brainplane's relay, end-to-end encrypted. The relay cannot read them.

It reads only what it needs to track work:

• Codex and Grok session lists, read-only: session ids, folders and times, to link a session to a topic and build the resume command.
• Hook events: session start, a new prompt, tool use and the end of a turn. On a new prompt, brainplane uses the prompt to find the topic and add its brief. When an agent waits, the card keeps the folder name and the agent's last line.
• Files the agents touched, by path.
• brainplane never opens your transcript files. To draft a new skill, it asks the session's own agent to read its history.

No. The app has no analytics.

The website counts visits with Google Analytics only if you accept it in the cookie banner. Before you choose, and if you refuse, nothing loads from Google. No advertising. My account sets 1 sign-in cookie, only after you sign in. The waitlist keeps your email, the date you joined and which form you used. To block repeated sign-ups, the server reads your IP address in memory for a few minutes and does not save it.

brainplane checks every write against a list of known secret formats. When it recognises one, it refuses to store it and reminds you to change it. Strings that look like bank account numbers are masked. Its own logs keep ids and sizes only.

No check catches every secret, so keep passwords and keys out of your notes.

Yes. Each topic belongs to a scope, such as work or family, set by the folder it lives in. A list per vendor says which scopes its agent may receive. A topic with no scope goes to no agent.

You change the folder map and the vendor lists in brainplane's settings file.

Your data is plain files in your brainplane folder: copy it to keep it, delete it to remove it. Uninstalling keeps it until you delete it.

$12 a month, or $99 a year (31% off). Prices are in US dollars, tax included where tax applies. Paid checkout is not open yet. The 14-day trial works today.

The price covers brainplane only. Your Claude, ChatGPT or Grok plans are separate and paid to those vendors.

Yes. 14 days with every feature. Only an email address, no card. 1 trial per email and per Mac.

brainplane goes read-only. Your data is never locked and never deleted.

• Keeps working: reading every topic, decision, receipt and card; answering cards; export; the MCP read tools, so running agents do not break.
• Stops: new handoffs, Autopilot and standing answers.

Yes. Ask within 14 days of your first payment, monthly or yearly, and you get a full refund, no questions asked. Write to hello@brainplane.app. Your legal rights as a consumer apply on top of this.

From My account on brainplane.app: sign in with your email, then Manage billing. Your plan runs until the end of the period you paid for, then brainplane goes read-only.

$69 a year for the first 500 yearly buyers. The price stays while your plan renews. The 500 seats are counted on our server and the cap does not move.

No. Agent tools change every month, so the work to keep up never stops. When a plan ends, brainplane goes read-only and your data stays readable and exportable.

1 person, up to 3 Macs. Students and teachers get 50% off the yearly plan for the first year.

Yes. Open app.brainplane.app in Safari, add it to your Home Screen and pair it with your Mac by scanning a code. The lock screen says an agent needs you. Open the app to read the question, end-to-end encrypted, then hold 1 second to answer, with 5 seconds to cancel. It also works on Android and in any browser.

The app runs on the Mac. Agents on Linux servers, x64 or arm64, join it with bp node, over SSH with your own keys and with no open port on the server.

Yes. A licence covers up to 3 Macs. Pair them and each Mac shows the other's agents and cards. When 1 Mac sleeps, requests from your phone go to a Mac that is awake.

brainplane is built for 1 person: 1 licence per person, on up to 3 Macs. For a team plan, write to hello@brainplane.app.

Those are free tools to run many coding agents side by side in a terminal. brainplane works on what your agents know and on when they need you: shared context, handoffs, standing answers and 1 Inbox. You can use them together.

Write to hello@brainplane.app. Add your brainplane version (in Settings) and the output of bp doctor, which checks the local service. It prints version numbers, the path of your brainplane folder and whether the service runs.

The brainplane team builds it. The company that publishes and sells brainplane is named in the legal notice, linked at the bottom of every page.