---
title: Changelog — what changed in instibrain, and when
description: Every change a person meets in instibrain, newest first and dated: the cockpit, each release of the sync client, the brain's tools, sign-in, and the system's rules.
url: https://instibrain.ai/changelog
updated: 2026-11-12
site: https://instibrain.ai/llms.txt
---

# What changed, and when.

One entry for each change you meet: in the cockpit, on your computer, through your AI, or at sign-in. Newest first, each day dated, each release of the sync client numbered.

Sign up free  →  Take the tour

237
changes a person meets, since 26 October

72
releases of the sync client

0.0.110
the newest release; your computer takes it at its next daily update

16 Nov
the newest change

the log

## Every change, newest first.

Each row names the part it changes: the cockpit at my.instibrain.ai, the sync client on your computer, the brain your AI reads, your account, the system every AI reads first, or the installer. A row with a number is a release of the sync client.

-  16 November  2026
-   sync client

###  0.0.110 A slow brain never holds your prompt, and no room message is lost

Before each prompt, the sync client reads the new messages in your window's rooms. It ends within six seconds of its start, however slowly the brain answers, and Claude Code waits ten seconds for it. Each call to the brain goes once, and starts only when it can end in time. A room it does not reach waits for your next prompt, and a message it read always reaches the window. A shorter wait in your Claude Code settings becomes ten seconds when your next session starts.

-  13 November  2026
-   sync client

###  0.0.109 Claude Code stays on the newest model at xhigh effort through a model release

The setup names no model's version, so a model release needs no change on any computer. Claude Code runs at the xhigh effort level on every model, a new one too: the setup holds the level in Claude Code's CLAUDE_CODE_EFFORT_LEVEL setting, where Claude Code alone starts each new model at medium. That level decides over /effort; a value you set there yourself stays, and your AI opens a window at another level when you ask. A model picked in Claude Code's model menu is kept by its family, opus, sonnet, haiku or fable, so it follows that family's newest model. The menu offers today's models: a model list Claude Code has not renewed for a day is removed, and Claude Code fetches it anew. Each update applies these settings, on a new computer and on yours.

-  12 November  2026
-   sync client

###  0.0.108 An open AI window takes up an update with no restart

After an update, an AI window opened on sync client 0.0.108 or newer takes up the new client by itself: its instibrain MCP looks at most once a minute for another client in place and hands each call to it, so the brain hears the new version and the window needs no restart. If the new client stops answering, the call it held gets an error and the window's own client serves again. A window opened on an older client keeps that client until it closes. The update ends each rooms watch from before it, on Windows too when Git Bash started it, and the session starts its watch again from the new client. One reach of a Claude limit is one notice: a window that still reads under 100 beside one at 100 no longer counts the reach again.

-   sync client

###  0.0.107 A space no longer shared with you leaves your computers

When a person stops reading a space, by their own share, a group's or Everyone's, the sync client removes the space from each of their computers at its next sync: its pages, their conflict copies, its link and its keys, and the git history the client made. A file changed there and never sent moves to ~/.instibrain/kept/. The client says what it removed, and the next session's start says it again. A watch stops within five minutes. A person who leaves the account, and an agent's computer when its share ends, leave the same way. A client older than 0.0.107 keeps its files until it updates.

-  11 November  2026
-   sync client

###  0.0.106 No Terminal window opens on Windows for a hidden job

On Windows, the jobs that run out of sight, the session-start check, the limits hook and the background watch, open no Terminal window when they unseal the computer's keys, read the dark-mode setting, or run the update.

-   sync client

###  0.0.105 The brain's files read under their own names and the names they had

The sync client reads identity.md, index.md, user-profile.md and rules/working-here.md under those names and under the names they had, so a space works before and after its pages are renamed. It writes CLAUDE.md and AGENTS.md at ~/instibrain, with the same text, after each sync. A sealed space takes the new names on a computer that holds its key: each file moves and keeps its history.

-  10 November  2026
-   sync client

###  0.0.104 A rooms watch ends with the session that started it

instibrain rooms watch stops when the Claude Code session that started it is gone, so a closed session on Windows leaves no watch behind. The watch reads the computer's process list once, not every 15 seconds, and Windows looks up a session's window once, not at every prompt.

-  9 November  2026
-   sync client

###  0.0.103 What the brain lost comes back from your computer; specialists kept, keys sealed, releases signed

After a brain restored from a dump, the sync sends back once what this computer holds that the brain lost, and refuses the brain's older version of a page, so the file here keeps its text. A specialist the brain no longer lists stays whole on the computer. The computer's private keys sit sealed in its files, so a copy of a home folder opens no sealed space. Each release carries a signed manifest that expires, and the installer checks it.

-   brain

### The service was down from 10:46Z to 12:45Z, and came back from the morning's dump

A check's cleanup removed production's data at 10:46Z. Both databases came back from the 08:00Z dump, with who may reach what as of 10:45Z, so nothing granted since the dump came back. Pages written between 08:00Z and 10:45Z were lost on the server, and came back from a computer that held them where one did. What protects a restore now: the data stands outside every check, a check runs on a daemon of its own and is removed by its label alone, the brain tells a computer's sync that it was restored, and the key that opens open pages at rest has a copy off the server.

-   brain

### Search finds a page by its words and by its meaning

Your AI's search finds a page by its words and by its meaning, in one list, the best first, across the open spaces you read. A question in other words than the page's, or in another language, finds it too. The meaning side runs on the bge-m3 model, chosen on a fixture of this repository's own docs. The brain keeps the vectors it searched in memory, so a search of a large space opens only what it lacks, and the worker indexes new pages on four threads at once.

-  8 November  2026
-   account

### Two-step sign-in holds against a stolen password

Ten wrong second-step answers in fifteen minutes close the step, and the first wrong answer mails you that your password was used. Setting your first second step signs out your other sessions.

-   sync client

###  0.0.102 Steadier hooks in your Claude Code windows

A window waits at most three seconds to hear its rooms, then moves on, and shows each message's day. The save guard reads more ordinary words as prose, such as a mail address or a setting's name.

-   system

### The Keeper page guards against empty secrets

The Keeper tool page tells your AI how a lapsed Keeper sign-in shows, to stop when a secret comes back empty, and that a JSON read shows every field's value.

-   sync client

###  0.0.101 The writing check names history in a page

The writing check on a saved page names what happened rather than what is: a dated action, a phrase of history such as "no longer", a dated note, and a rule that only forbids.

-   sync client

###  0.0.100 Windows follow your computer's light or dark look

instibrain's Claude Code windows follow your computer's light or dark appearance. On a Mac, `instibrain window --recolour` gives each open iTerm2 window the colours of the current appearance, each window keeping its own colour.

-   system

### The rules every AI reads open with ten lines

The pages every AI reads first open with ten lines it keeps in every window, such as answer first and short, and wait for your go before a send. A page holds what is true; what happened lives in operations, the inbox and rooms.

-   cockpit

### Lists take filters beside their tabs

Each list's tabs answer one question, and other filters sit beside the search as boxes that combine with them, so Assigned to me on Operations combines with any state. A filter that leaves no row shows one Clear filters.

-  7 November  2026
-   cockpit

### Link a computer for a team agent

An owner or an admin links a computer for a team agent on Agents. That computer keeps the open spaces shared with the agent as files, and its AI works as the agent, at the agent's role. No person's Devices lists it.

-   cockpit

### The cockpit reads well with a screen reader

Each page carries headings and landmarks a screen reader follows, and every chart is named by its card. The browser tab names the page and your unread count, such as (3) Spaces · instibrain.

-   cockpit

### Lists show times and roles one way

Every list shows a moment as how long ago, such as 7h ago, with the exact moment on hover, on a 24-hour clock. A role is a word in the row, and a room's header says its kind: Direct, Group or Operation.

-   cockpit

### Each page opens as you left it

Each page of the cockpit opens on the tab and the view you left it on, remembered in your browser, Agents' Cards or List included. Settings' tabs have their own addresses, as People's and Devices' do.

-   cockpit

### The reach sheet says what your AI reads

The sheet of an AI's spaces is titled What <name> reads, and a space it does not read shows Not given.

-   sync client

###  0.0.99 Windows hear edited and deleted messages

Your Claude Code windows hear an edited or deleted room message, marked as such. A folder whose sync stops starts no sync on a later change, and the note that your account is at its limit reads as two clean sentences.

-   brain

### A team agent connects from Claude Code

A team agent's connect link names the agent in its address, so Claude Code signs in as that agent. The approval page says the login acts as the team's agent, and only an owner or an admin approves it.

-   cockpit

### Devices has three tabs

Devices opens on three tabs, each with its own address: Computers, AIs, and Limits and backups. Computers lists the linked ones first and the unlinked ones under their own pill, and each tab has columns that fit what it lists.

-   brain

### Edit or delete what you sent

You and your AI edit or delete your own message in a room, and its readers see it marked edited or deleted. Your AI edits or withdraws an inbox item it sent while it waits, and the cockpit offers Undo.

-   cockpit

### The palette offers an action when you name it

The Ctrl K palette offers New page and Share only when your words name them, so typing Devices and pressing Enter opens Devices.

-   cockpit

### The keyboard keeps your place

Closing a dialog, a sheet, a menu or a rename returns the focus to the button or the row that opened it, and Undo in a toast returns it where it was. Enter on a person's row of Activity opens their page.

-   cockpit

### Undo a proposal's Withdraw, Clear or Reject

Withdraw, Clear and Reject on a proposal take its row away at once, with Undo in the toast; the change goes through when the toast closes.

-   cockpit

### Shared with opens the sharing sheet

For a manager, a click on Shared with on Spaces, Specialists or Operations opens the Sharing sheet, on a long row too. A click on a cell of people, a role or a state never opens the row.

-   sync client

###  0.0.98 instibrain check accepts inbox.md at the root

`instibrain check` accepts inbox.md at a space's root on every template, as the server's check does.

-   brain

### The structure check accepts inbox.md at the root

The structure check counts inbox.md at a space's root as a reserved page on every template, as CLAUDE.md and MEMORY.md are, and never names it a loose page.

-   brain

### Rooms reach only a window that is open

When your AI sends to a window or adds one to a room, the window must be open on a live computer; otherwise nothing is sent, and your AI hears the open windows, the closest name first. A window's name matches without its icon.

-   cockpit

### Operations, Devices and Activity fit a laptop screen

Operations, Devices and Activity fit their card from 1280 pixels wide, with no sideways scrolling; a long name is cut and reads whole on hover. On Activity, a person with many computers keeps one row: two computers show and the rest are counted.

-   sync client

###  0.0.97 The curator gives each operation its emoji

The curator's daily run and `instibrain check` name each operation of an open space with no emoji. The run sets the emoji its index line shows, or suggests one in its report.

-   brain

### The structure check names an operation with no emoji

`check_structure` names each operation of an open space that has no emoji. The AI that starts an operation sets its emoji with its brief, and the curator's daily run sets the emoji the operation's index line shows.

-   sync client

###  0.0.96 The desktop menu finds each operation's emoji

The desktop button's menu shows each operation's emoji and assignment when a space's folder on your computer has a different name from the space.

-   sync client

###  0.0.95 The desktop menu lists your own operations

The desktop button's menu lists only your operations: in an open space, those assigned to you; in a sealed space, every one, as only your computers hold them.

-   brain

### An operation is assigned to the person who starts it

The person whose page starts an operation, from their AI or their computer, is assigned to it and joins its room. Each operation of an open space with nobody on it takes the person who wrote its first page.

-   cockpit

### Operations read the same in the cockpit and the terminal

Operations shows each operation's name as words, capitalised, as your computer's windows do, and its search still finds the folder's name. For an editor, a click on Assigned opens the Assign sheet and a click on the emoji opens the emoji picker.

-  6 November  2026
-   brain

### One message reaches several people at once

A message goes to several people and agents at once, into one room that keeps exactly that set and is found again whoever sends. In the cockpit, People's + starts a talk with several, listed first.

-   cockpit

### A member's Home shows the account at a glance

A member's Home shows the account, its computers and what changed, with no week of activity, which only owners and admins read.

-  5 November  2026
-   system

### Your AI works with Gmail stars, labels and filters

Through the Google Apps Script tool, your AI stars messages, creates and applies Gmail labels, and lists, creates and deletes filters. The script says which permissions still wait and gives the link that grants them, Workspace accounts included.

-  4 November  2026
-   brain

### A proposal never loses a fact

The brain refuses a proposal that removes a page while one of its facts stands on no other page. Your no to a proposal of the curator's daily run is kept, and the run never proposes it again until an editor lifts it.

-   sync client

###  0.0.94 The curator's daily run keeps every fact

The curator's daily run proposes only cuts that keep every fact: filler words, and sentences another page already holds. It never proposes a move, a split, a merge or a removal; it names each for you instead.

-   sync client

###  0.0.93 One release key, and no secrets in sealed pages

Your computer takes updates signed by one release key alone. A sealed page that holds a secret, such as a password, stays on your computer until the secret is out, however the page was saved.

-   brain

### Your AI finds a teammate by a first name

A tool that takes a person finds them by one word of their name, such as a first name, when no whole name matches. A word two people share names both.

-   cockpit

### Activity counts Codex and the curator's run apart

Activity counts Codex sessions beside Claude Code's, keeps the curator's daily run apart from a person's own work, and names the Claude plan behind each limit. The `activity` tool answers a figure it does not know as unknown, never 0.

-   sync client

###  0.0.92 Signed writes, and keys in your computer's key store

Your computer signs every write it sends and locks its keys in its own key store, such as the Mac's Keychain. A teammate's new computer joins a sealed space once they approve it on a computer of theirs, and Codex counts on Activity.

-   brain

### A copy of the database reads no page

The brain stores every open page's text, versions, words and path under a key it keeps outside the database, so a database login or a copy of the database reads no page. The brain itself still reads them, to serve and to search.

-   cockpit

### The read trail covers specialists

Reads, on the Brain page, lists each specialist beside the spaces: who opened its pages, an old version, or took it whole. Its managers see every read; anyone else who reads it sees their own.

-   cockpit

### A removal says what the person keeps

When you take someone off a space or out of the account, the cockpit says that what they already read or synced stays with them.

-   brain

### Confidential text waits for a manager before it spreads

After your AI reads a confidential or restricted page, for an hour its writes into a space with a wider label wait for a manager. The manager reads the text and lets it land in the cockpit, or through their AI on their yes.

-   brain

### Labels say how far a page's text may travel

Every space and page carries a label, public, internal, confidential or restricted, and every read names it to your AI. A manager sets it in the space's Settings, and a share wider than the label allows is flagged, never stopped.

-  3 November  2026
-   brain

### One inbox across the spaces only you read

Your inbox is the inbox of every space only you read, and the inbox tool lists it as one list. It also names each other space you manage whose inbox holds items.

-   system

### Your AI writes enough words to carry the intent

instibrain's rules teach your AI to use enough words to carry the intent and none that only add noise, then to reread as its reader before it sends or saves.

-   sync client

###  0.0.91 Releases are signed on a computer apart from the server

Releases of the sync client are signed on a computer apart from the server. This release teaches your computer the new signing key, inside a release the old key signed.

-   cockpit

### An inbox item closes with Done

Each inbox item in the cockpit has one action, Done, with Undo. Filing an item into pages is your AI's work, on your computers.

-   account

### See what staff did on your account

A password, invite, or new owner's link goes to the person's mailbox alone; staff never see it. Settings ends with Staff acts on this account: owners and admins see every act, a member the acts on themselves.

-   cockpit

### See who read a space's pages

A space's Overview lists its reads under Changes: who read which page, through what, and when; a manager sees every read, anyone else their own. Activity shows each person's pages read as counts, with each burst marked.

-   brain

### Every page read is recorded for 30 days

Each time a page's text goes to a computer, an AI, an agent, or the cockpit, the space records who read it and when, for 30 days. One caller opening many pages within ten minutes mails the space's managers and the account's owners and admins.

-   brain

### A computer asleep over a weekend is not behind

Owners and admins hear of a computer whose daily update is two days late only once it is back in use and has had an hour to update. A computer asleep over a weekend raises nothing.

-   sync client

###  0.0.90 Claude in Chrome loads in instibrain's windows

Claude Code windows that instibrain opens take your Claude in Chrome settings, so the browser tools load there and the Chrome introduction does not show again.

-   cockpit

### Activity shows work outside instibrain

Activity's List adds two columns for Claude Code sessions and prompts in other folders, with a dash where a computer declined. An active hour counts a prompt anywhere, once, and someone who works only elsewhere still shows as having worked.

-   sync client

###  0.0.89 Work in other folders counts unless you decline

Your computer counts Claude Code sessions and prompts in other folders, as numbers alone, unless you decline with `instibrain setup decline elsewhere`. It reports your Claude limits from any folder, and on Linux the daily update runs at ten, as on a Mac and Windows.

-   cockpit

### Activity shows when each person was last seen

Activity's List says when each person was last seen, under their name. Each computer's Claude limits show its highest use, or a dash with the reason it read none, such as no Claude sign-in.

-   cockpit

### Activity counts sessions and reads alike everywhere

Sessions on Activity, Home, and the year grids count Claude Code sessions alone, and a session open for days shows its prompts and hours. The check each window makes for new room messages counts as no read.

-   installer

### The installers name where to make the link code

The installers say where to make the link code that joins your computer: the cockpit's Devices page.

-   sync client

###  0.0.88 The daily update waits for the network

The daily update tries a failed download again until the network is back, as after a Mac wakes offline. On Windows, Git Bash runs the instibrain command the hooks call, and the setup check keeps the hook that lets a window hear its rooms.

-   cockpit

### Agents shows each agent as a card

Agents shows each agent as a card, with its Settings and Details a click away, and keeps the table as its List view. A link that names an agent marks its card and scrolls to it.

-   account

### Every sign-in takes a second step

After the password, you confirm with a passkey, an authenticator code, or a code by mail; someone with none sets one up before anything opens. The Security tab in Settings adds, names, and removes each. A sign-in from a new browser mails you.

-   cockpit

### Activity and Devices open on charts

Activity's Overview charts how the desktop button opened sessions, fresh, resumed, as a specialist, or on an operation, and sets one person against the team's median. Devices' Overview shows each computer's Claude usage limits and how many backups are up to date.

-   cockpit

### Large spaces show every page

A space's map shows every page, up to 5,000, and a page's history every version, up to 1,000. Changes filters by who wrote, and Show more reads further back in each space.

-   cockpit

### Spaces and Operations show who they are shared with

Spaces and Operations gain a Shared with column, as Specialists has: a manager sees who, anyone else how many people read it. Each column keeps one name across the cockpit, such as Last seen, Pages, and Reaches.

-   sync client

###  0.0.87 Devices shows each computer's backup

Your computer tells instibrain when it saves a sealed space's recovery bundle. Devices shows that time beside the computer's newest key, marked Out of date when a key came after, or No backup.

-   sync client

###  0.0.86 A sealed space is reviewed before a new reader's key

Before a sealed space's key reaches someone newly shared with it, a manager's computer checks its pages by the rules an open page others read must meet. It holds the key while a finding stands and names each page in its sync report.

-   sync client

###  0.0.85 The curator proposes to dissolve a mixed page

The curator's daily run proposes to dissolve a page that mixes three kinds of line or more: the proposal says where each line belongs and removes the page once an editor accepts. `instibrain check` names mixed pages too.

-   brain

### Your AI hears when a page mixes kinds of line

When a save brings a log, a list of decisions, or a customer's id to a page of another kind, the brain names each line and where it belongs, and saves the page all the same. The structure check names them on every page.

-   brain

### A specialist always keeps a manager

Taking a person out of a group, or deleting the group, is refused when that group alone manages a specialist, as it is for a space.

-   cockpit

### People shows each person's computers and last sync

An owner or an admin sees, for each person, how many computers they run, which are behind and on what version, and their newest sync in your time zone. Pills filter people by role.

-   cockpit

### Every list sorts by any column

Spaces, Specialists, Operations, People, Groups, Agents, Devices, and Activity sort by any column from its header, with a hint saying what it holds. A filtered list says how many rows it shows of how many, with Clear filters, and its address keeps the view.

-   installer

### The Linux installer marks the desktop buttons to open

The Linux installer marks the two desktop buttons as trusted programs, so GNOME opens them on a double-click from the first run.

-   sync client

###  0.0.84 Linux desktop buttons open on a double-click

On Linux, the daily update marks the two desktop buttons as trusted programs, so GNOME opens them on a double-click.

-   cockpit

### Home shows your team's week

Home opens on six figure cards: what waits for you, today, and the computers behind, each opening its page. Charts of the week follow: sessions, prompts, pages written, reads, active hours, the busiest people, and the newest changes. A member sees their own week.

-   cockpit

### Rooms lists people and AIs apart

Rooms splits its direct talks in two: People, every person who joined, by name; and AIs, every agent, then your own named windows. Each list folds and sums its unread messages, and the search finds a talk in either.

-  2 November  2026
-   cockpit

### Activity's Overview compares your days with the days before

Activity's Overview sets sessions, prompts, active hours, and pages written beside the same days before, draws each day's sessions by the busiest five people, and shows the year's usage, contributions, and session starts as grids. A person's page shows their own grids.

-   cockpit

### Owners and admins edit a person's name on People

On People, an owner or an admin picks Edit name in a person's row menu to change their full name, joined or invited; only an owner renames an owner. The person gets no mail about it.

-   cockpit

### Home's Waiting for you counts what the bell counts

Home's Waiting for you starts with the bell's number and See all, then a line per room with unread messages, a line per space with proposals, conflicts, restored pages, and inbox items, and the computers on an older sync client.

-   cockpit

### Set your time zone in Settings

Settings, on its Account tab, takes your time zone as a city, such as Lisbon, or fills in this browser's zone. AIs that run outside your computers, such as Claude on the web, say times in that zone.

-   sync client

###  0.0.83 Your computer tells instibrain its time zone

Your computer sends its time zone with every call, so your AI says each time in your zone. Each window also names its session, so a message to you reaches the window of the specialist you picked in Settings.

-   cockpit

### The bell lists what waits for you

The bell lists what waits for you: messages for you, proposals to close, items sent to your inbox, sync conflicts, and restored pages; Notifications shows them all. Owners and admins also see joins, linked computers, connected AIs, failed setups, and Claude limits reached.

-   brain

### Messages your AIs send are not unread for you

In the cockpit, a message you or one of your AIs sent, from any window, never counts as unread for you. Your windows still count each other's messages, so two windows of yours work together in an operation.

-   brain

### One of your windows answers each message to you

A message to you, a group of yours, or your space wakes one open window of yours, once: the window of the specialist you pick under Messages to you in Settings, else the window that spoke last.

-   sync client

###  0.0.82 Every linked folder keeps its background sync

On Windows an update restarts the background sync on the new version, and on Linux every folder you link gets its own. `instibrain window` takes --model and --effort, and your computer reports how each session began and your Claude limits to the cockpit.

-   system

### Your AI says each time in your time zone

Your AI says every time in your own time zone, never in UTC, and an AI on your computer takes that zone from your computer's clock.

-   brain

### Spaces from a retired template pass instibrain check

A space made from a template instibrain does not offer any more, such as team or software-project, carries an empty template, so `instibrain check` checks that space like any other.

-   cockpit

### Activity counts days in your time zone

Activity counts each day in your browser's time zone, so today, the dates you pick, and the year grids all follow your own clock.

-   cockpit

### Activity lists each person's days and computers

On Activity, owners and admins see each person grouped as worked, reads only, nothing yet, or invited, with sessions, prompts, active hours, pages written, what they opened, and each computer's installed version, marked Behind when it trails the release.

-  30 October  2026
-   cockpit

### Inbox items and search lines show their formatting

Inbox items on the Brain page and the lines Search finds in a page's text show bold, code and links as formatted text. A link in an inbox item opens its page in the space.

-   cockpit

### Star the spaces and specialists you use most

Star a space or a specialist from its row's menu on Spaces or Specialists. Every list, Home's Waiting for you included, shows your starred ones first, on every computer; the star is yours alone.

-   brain

### Every space has its room

Each space has a room that everyone it is shared with hears, under the space's name, and every share or unshare keeps its members in step. Your AI hears it only where it reaches the space; the cockpit lists these rooms under Spaces.

-   brain

### A new space is yours alone until you share it

A new space starts with its maker as its only manager, from the empty template: a CLAUDE.md with its purpose, and MEMORY.md. Everyone reads it once a manager shares it, after the share's review; New space asks only whose it is.

-   brain

### Erasing a space's history takes write

Your AI erases a space's history only where you gave it write; an AI that only reads is refused and told how you give it write. General always keeps Everyone: no owner or admin removes the group from it.

-   sync client

###  0.0.81 The curator settles structure findings each day

The curator's daily run takes each space's structure and share findings and settles or proposes each; `instibrain check` accepts links into the system. A computer stopped between taking a key and the end of its cycle pulls its pages again from the start.

-   cockpit

### The Brain page shows only what needs you

Waiting for you on the Brain page holds inbox items, conflicts, restored pages and proposals. What your AI handles itself, such as structure and share findings, is one quiet line under it, each kind once with its count.

-   sync client

###  0.0.80 Open a Claude Code window on a server

`instibrain window --ssh` opens a Claude Code window on a server, in your colours and theme, inside tmux when you name a session. It reconnects after a dropped connection and types nothing into a session that already runs.

-   installer

### Windows starts the sync client's watch at sign-in

The Windows installer makes the shortcut that starts the sync client's watch at each sign-in once a folder is linked, upgrades included. Its closing check names the shortcut, and the reason when it was not made.

-   cockpit

### A list's filter, search and page stay in the address

A list's pill, search and page live in the address, so Back, a reload, and a shared link show the same rows.

-   cockpit

### Forms say under each field what to type

A form's button stays ready until you press it; a field that fails says under it what to type and takes the focus. A refusal about one field shows under that field, and Save with nothing changed simply closes.

-   cockpit

### Home shows what waits for you

Home answers what waits for you, what changed lately, and the next step. Waiting for you lists each room with unread messages, each space with inbox items, conflicts, proposals or restored pages, and computers behind; each line opens where you answer it.

-   cockpit

### The cockpit follows your reduced-motion setting

When your computer asks for less motion, the cockpit's animations end at once, the loading line stands still, and a jump scrolls at once.

-   cockpit

### A page under another links back to its parent

A page opened from another, such as a person's Activity page, shows a link back to its parent above its title.

-   cockpit

### Red, amber and green mark a state alone

The cockpit's red, amber and green appear only for a state; a sealed space, a role, an agent, or a kind of change takes a calm colour of its own. Unread lines are amber: they wait for you.

-   cockpit

### Sheets and dialogs fit every screen

A long sheet keeps its Save button in view at its foot, even over a list of twenty people. Dialogs keep a margin on a phone, with their buttons at the right at every width.

-   cockpit

### The switcher shows every space, Sealed included

The space switcher is wide and tall enough to show every space by its whole name, and offers New space when there is none. New space opens as one sheet, with the same fields, from the switcher and from Spaces.

-   cockpit

### Lists, rows and menus read alike

Lists load as skeletons of their rows, an agent shows its emoji where an avatar sits, and a filter no row matches turns grey instead of a red zero. Devices marks only Idle and Revoked devices; Brain page menus keep each item on one line.

-   system

### Your AI replies inside a Gmail thread

The Google Apps Script tool drafts a Gmail reply inside its thread and sends a draft on your yes, so your AI answers where the conversation is. Gmail search pages past 50 results, and marking read or trashing takes a whole thread.

-   sync client

###  0.0.79 Windows wake to their rooms, direct rooms stay direct

A window opened with no task wakes at once and starts listening to its rooms. A direct room between two windows reaches only those two, never the person's other windows.

-   cockpit

### Computers behind the current sync client show first

Devices shows each computer's sync client and last update, marks it Behind when it is older than the release, and says how to catch up. Home lists computers behind first, everyone's to an owner or admin, and Activity lists each person's computers.

-   brain

### See which sync client each computer runs

`list_computers` and the activity tool name the sync client each computer runs, its last sync and update, and whether it is behind the current release. A member sees their own computers; an owner or admin sees everyone's.

-   cockpit

### Every removal in the cockpit has an Undo

Removing a share, a person's recovery of history, a row, or a specialist's page hides it at once, and the toast offers Undo until it closes. ⌘Z or Ctrl+Z undoes the newest removal while its toast shows.

-   system

### Windows your AI opens keep your colours

A Claude Code window your AI opens by hand takes the profile, colours and theme the sync client prints. A server's window runs in tmux only where the server has it; on a Mac your AI opens a new window, never typing into yours.

-   brain

### Rooms for your AIs, each operation, and one task

Your AIs is a room for you and every AI of yours. An operation's room holds its assigned people and ends with the operation; a temporary room an AI opens for one task ends 10 days after its last message.

-   sync client

###  0.0.78 Your Claude Code windows hear their rooms

A Claude Code window hears its rooms with nobody typing into it: new messages arrive in its next turn, and a waiting window wakes when one comes.

-   system

### Every AI works as a teammate in rooms

Every AI reads that a room is where AIs talk while their people read along. At wake it reports unread rooms; it asks whoever holds what it needs, tells whoever waits, keeps secrets and sealed facts out, and puts what matters on a page.

-   brain

### See who is working now in each room

`list_rooms` names who is working in each room right now: each window, AI and agent that hears it and acted in the last 10 minutes. Each room tool tells your AI in one line how to work as a teammate.

-   installer

### The installers check the sync client's signature

The installers for Mac, Windows and Linux install the sync client only when its checksum and its signature by instibrain's release key both match; otherwise they stop and say why.

-   sync client

###  0.0.77 Updates install only a signed sync client

The daily update installs a new sync client only when its checksum and its signature by instibrain's release key both match.

-  29 October  2026
-   cockpit

### The tab counts your unread messages

The browser tab's title counts unread messages across every room, as (3) instibrain. Ctrl+K puts the best match first, a room before a page for a room's name, and Enter opens it.

-   cockpit

### Your windows appear on Devices from their first call

A named Claude Code window shows on Devices and in the Rooms screen's Direct list from its first call, a read included, until 24 hours pass without one.

-   sync client

###  0.0.76 A window's name comes from Claude Code itself

Your computer's AI reads its window's name from Claude Code's own record, so a /rename shows within seconds. One update runs at a time, and a failed one stops cleanly and records why.

-   cockpit

### Rooms in the cockpit, laid out like a messenger

The Rooms screen lists your rooms and a direct talk with every person, agent and named window of yours, unread ones in bold. Open one to read and send messages, manage its members, or keep a message as a page.

-   sync client

###  0.0.75 Each Claude Code window keeps its own place in a room

Your computer's AI names the Claude Code window it serves on every call, so each page and message records its window, and each named window keeps its own place in a room.

-   brain

### Rooms: where people, agents and AIs talk

A room is where your team talks: General for everyone, group rooms under a name, and direct rooms. Every AI of yours hears your rooms as you and sends messages there; a message holds up to 8 KB and is gone after 30 days.

-   cockpit

### The Activity page shows a year of each person's work

Owners and admins open Activity for a year of usage and contributions as a grid of days, for the account and for each person, with what each counter means. Members do not see it.

-   cockpit

### The cockpit gains Agents and Devices pages

Agents lists the team's agents, what is shared with each, where it runs and its last use; owners and admins make, connect and remove them. Devices lists each of your computers and AIs with what each reaches.

-   brain

### Agents belong to the team

An agent, such as a support bot, belongs to the account: an owner or an admin makes it and approves each of its logins. A manager shares a space or a specialist with it after the review a person gets; it reads no sealed space.

-   sync client

###  0.0.74 Your computer counts sessions and prompts for Activity

Each session and prompt in ~/instibrain, and each pick from the desktop menu, counts toward your activity, sent in a ping of its own that never delays a prompt and holds nothing of what you typed.

-   brain

### Owners and admins read the account's activity

The brain counts each person's use of instibrain, sessions, prompts and AI calls, and what they add for others, pages and proposals, by day. Owners and admins read it with the `activity` tool; a count never holds a prompt or a page's text.

-   brain

### Each AI you connect acts as you, within its reach

An AI you connect, such as Claude on claude.ai, acts as you in the spaces and specialists you give it, to read or to read and write, never above your role. No space's sharing lists it, and a sealed space stays out of its reach.

-   cockpit

### Each space's sharing sheet replaces the Sharing page

Who a space is shared with lives on the space: its sharing sheet opens from its row on Spaces, from a page on Brain, or from the palette. People and agents sit on separate tabs, and each row picks its role.

-   system

### Each session reads your row on the team page

At the start of a session, your AI reads the team page of each team space you read, finds your row, and knows what your week and month deliver.

-   sync client

###  0.0.73 A window set to the default colour says so

A window opened with the default colour reads Claude Code's answer and reports its colour set after one try.

-   cockpit

### New space asks what the space is for

New space asks what the space is for and makes that purpose its first page; your AI's `create_space` asks too. A new space starts as a working space, without the profile and owner's layer that belong to Personal.

-   sync client

###  0.0.72 A window opened on a Mac starts

A window opened on a Mac with `instibrain open --here` starts its session with the settings it needs.

-   sync client

###  0.0.71 Your computer tells the brain its last update

Your computer tells the brain when its daily update last succeeded, so a computer that falls behind is seen.

-   account

### Owners and admins hear when a computer stops updating

When a computer that synced in the last two days has gone two days without a successful daily update, every owner and admin of the account gets a mail, once each time it falls behind.

-   sync client

###  0.0.70 Sessions in ~/instibrain load only instibrain's setup

A session opened by the button, in ~/instibrain, or by the curator loads only what instibrain installed, never your own Claude notes, rules, skills or plugins. `instibrain setup check` lists those, and `setup quarantine` moves them aside with your yes.

-   brain

### Managers see a space's history by their role

A manager of a space sees its history with no setting: each change and who made it, old versions, removed pages, and restore, through your AI and in the cockpit. Readers and editors see the space as it is.

-   brain

### Send an item to a teammate's inbox

Your AI puts an item in a teammate's inbox with `send_to_inbox`: one line in their Personal space's inbox.md, naming you as its sender. You read nothing of their space, and an item for yourself goes in your own inbox.

-   system

### Each inbox names who empties it

Your inbox is your Personal space's inbox.md: you and your AIs add to it and empty it. A team space's inbox.md belongs to its managers, and a new team space starts with one that says so.

-   sync client

###  0.0.69 Windows syncs from each sign-in; setup flags old sessions

On Windows, syncing starts at each sign-in and again after a failure, and an update waits for a program holding a file. `instibrain setup` flags a model pinned to an older generation and a Claude Code session started before its update.

-   sync client

###  0.0.68 The daily update catches up after a missed run

The daily update runs at ten, at each sign-in, and at the next start after a missed ten, and tries again when the network is down. Your computer links each space shared with you since, by itself.

-   system

### Every session starts by updating your computer

Each session on your computer starts by bringing the sync client up to date, and says so when it did. A Windows computer with a client older than 0.0.66 takes the install command again.

-   sync client

###  0.0.67 <code>instibrain --version</code> answers in any folder

`instibrain --version` prints the client's version in any folder. On Windows, `instibrain setup` repairs the link the client needs to start, and each update brings the daily update's task along.

-   system

### The first session sets your computer up

After its first questions, your AI updates the sync client, checks and repairs the setup, links your spaces and specialists, then connects your accounts, reusing a sign-in that already works before asking for one. It ends by reporting the setup.

-   sync client

###  0.0.66 Your computer reports its setup to owners and admins

`instibrain setup report` mails what your computer's setup came to, each step, space and connection, to the account's owners and admins. On Windows the daily update runs each day, or at the next start after a missed one, and keeps a log.

-   sync client

###  0.0.65 Windows Terminal on every Windows computer

The sync client installs Windows Terminal on every Windows computer, with no Store and no administrator, and the installer sets it up before Claude Code. In the old console, the button's menu opens in Windows Terminal, or draws as plain text.

-   brain

### On an owner's computer, your AI shares and invites

When you are an owner or an admin, the AI on your own computer shares spaces and specialists and invites people, each change shown to you first. It shares only a space you manage that the computer syncs; a member's computer gains nothing.

-   sync client

###  0.0.64 A window names and colours itself

A Claude Code window gives itself its name and its colour once it is idle, so nobody types /rename or /color by hand.

-   brain

### Editors change pages directly; anyone may propose

Your AI, as an editor or manager of a space, changes its pages and folders directly with your yes, new ones included, and indexes each new page. A reader, or anyone wanting a second opinion, asks for a change with a proposal.

-   sync client

###  0.0.63 Your computer files the specialists shared with you

Your computer keeps a folder only for the specialists shared with you; an owner manages the account's other specialists in the cockpit.

-   brain

### Specialist names are unique among what you read

A specialist's name differs only from the specialists its people read, not from every specialist of the account, so a teammate never hears that a name they cannot see is taken. A share that gives someone two of one name is refused.

-   brain

### A specialist's knowledge grows into topic folders

A topic that outgrows one page gets its own folder in a specialist's knowledge, with a hub page of the topic's name that lists its pages. Your AI makes one with `create_folder`, and the structure check accepts it.

-   brain

### The approval page lets you tick each permission

When you connect an AI, the approval page shows each permission as a box, ticked where the AI asked. You can withhold one it asked for, or give one it did not ask for, such as changing who reads what.

-   sync client

###  0.0.62 Each window wears its own colour

Each window takes one brand colour, pale in light mode and deep in dark mode: the home session sky blue, an operation orchid, the curator violet. A window opened on a Mac starts its session right away.

-   sync client

###  0.0.61 Each window follows your computer's dark or light mode

Every Claude Code window the sync client opens follows your computer's appearance, dark or light, whatever Claude Code's own theme says. The desktop button's menu fits inside its window and scrolls a long list.

-   brain

### Editors set a specialist's name, line and emoji

An editor or a manager of a specialist sets its name, its line and its emoji on the cockpit's Specialists page, or through your AI with `set_specialist`. Every emoji comes from a set every computer draws, so none shows as an empty box.

-  28 October  2026
-   sync client

###  0.0.60 Windows follow your Claude Code theme

Every window follows your Claude Code theme, light or dark, never the computer's. A dark theme gets a deep shade of the window's colour with light text, so no light text sits on a pale ground.

-   sync client

###  0.0.59 The curator's proposals carry the exact cut

A proposal the curator's sweep leaves names the exact lines to take out of the page, so any editor applies it as it stands. A page ranked for length alone gets no proposal.

-   brain

### An editor keeps a space's index files

An editor changes MEMORY.md and each top-level folder's index file, in the cockpit too, so a page they add gets its line. CLAUDE.md and the other skeleton files stay a manager's, and only a manager moves or removes one.

-   brain

### The account's owner manages every specialist

The owner of an account manages every specialist in it, with no share of their own, and finds each one on Specialists and on their computer.

-   sync client

###  0.0.58 A Mac window keeps its whole first message

A Claude Code window opened on a Mac starts with its whole first message, however long the task you hand it.

-   sync client

###  0.0.57 Windows open one at a time

Two window commands or menu picks close together take turns, so each window gets its own session and each command its answer.

-   system

### A server window outlives a dropped connection

A window your AI opens on a server runs its session in tmux, so a dropped network, a closed laptop or a closed window leaves the work running. The window attaches again when the connection returns.

-   sync client

###  0.0.56 Opus 5.5 runs at the xhigh effort level

Claude Code runs Opus 5.5 at the xhigh effort level by default, on a new computer and on yours at its next update. A level you saved yourself stays.

-   brain

### A proposal carries its change, ready to apply

A proposal holds the change itself: the full text of a new page, or the exact text to replace and its replacement. Any editor or manager has their AI apply it in one step, through the same save checks as any write.

-   sync client

###  0.0.55 Operations in the desktop menu

The desktop menu lists every operation of the spaces on your computer, under its own emoji, and opens a Claude Code window on it. The release also brings the curator's daily run and the writing check after each save.

-   sync client

### The curator tidies your pages every morning

Each day at seven, the curator ranks the pages on your computer that grew too long or drift from the writing standard. Your Claude Code then cuts them, merges conflict copies, and leaves a report your next session shows.

-   brain

### Every save is checked against the writing standard

After each save, your AI hears where the new text breaks the system's writing standard: a sentence over 30 words, a filler word, history on a page of state, or a sentence another page holds. The check reports and never refuses.

-   brain

### A proposal belongs to its space

Every editor and manager of a space sees each open proposal, in the cockpit and through their AI, and whoever makes the change closes it. A reader sees their own, and who answered it.

-   cockpit

### Every list in the cockpit goes by name

Every table and list of named things sorts by name, letters without case and a leading emoji ignored. Only a list whose order means something keeps its own, such as things that happened or a queue.

-   brain

### Your AI reads what you see in the cockpit

An AI you connect with read access sees what the cockpit shows you: who reads each space and specialist you manage, your people, groups, computers and sessions. Each tool tells your AI whether it reads or changes, and every list comes by name.

-   cockpit

### The Operations page

Operations lists every operation in the open spaces you read, the last changed first, with its space, its emoji and the people assigned to it. An editor sets the emoji and assigns the space's readers, there or through their AI.

-   cockpit

### A specialist has an emoji of its own

Pick a specialist's emoji in New specialist, or change it from its row's Emoji dialog: a grid of common ones, or any other. It takes the avatar's place on Specialists and comes first on the Brain page's map and in share sheets.

-   sync client

###  0.0.54 Each specialist's own emoji in the desktop menu

The desktop menu puts each specialist's own emoji before its name, in the menu and in the name of the window it opens.

-  27 October  2026
-   sync client

###  0.0.53 Menu sessions always take their colour

A session started from the desktop menu always takes its prompt colour, and waits rather than type over words you have started to type.

-   brain

### A specialist has an emoji of its own

A specialist carries one emoji of its own: your AI sets it when it makes the specialist, a manager changes or removes it, and your AI's list of specialists shows it.

-   brain

### Every operation in one list

Your AI lists every operation of the open spaces you read in one list, running and in planning, each with its space, brief, pages and last change, the latest first.

-   sync client

###  0.0.52 No conflict copy of your own text

Your computer never makes a conflict copy of a page it wrote itself. A conflict copy appears only when two writers changed the same page, for your AI to merge.

-   brain

### New tools reach a running session

A running session of your AI takes new instibrain tools as each release lands, with no reconnect.

-   system

### Server windows: SSH first, tmux for long work

A Claude Code window on a server opens over plain SSH; work that must outlive it runs in tmux, which survives a dropped connection or a closed laptop. Your AI watches it to the end and resends a message stuck in its input box.

-   brain

### Your AI does what the cockpit does

Your AI reads a space's recent changes and activity, a specialist's older versions, and your browser sessions. With your yes it restores a page, publishes a review, makes a specialist, sets a space's logo, grants recovery, erases history, and renames the account.

-   sync client

###  0.0.51 The desktop menu lists specialists by name

The desktop button's menu lists your specialists by name, and `instibrain spaces` lists your spaces by name, as every list of them does.

-   brain

### Your AI manages sharing, groups and people

With your yes for each change, your AI shares a specialist, renames a space, renames or removes a group, sends an invite again, signs a person out, and changes your own name and picture.

-   cockpit

### Spaces and specialists in name order

Every list of spaces and specialists, in the cockpit and for your AI, goes by name, ignoring case and a leading emoji, and a renamed row moves to its place at once.

-   sync client

###  0.0.50 Window colours arrive with every update

The sync client brings its iTerm2 window profile up to date each time it starts, so a computer updated from an older client gets the light palette too.

-   sync client

###  0.0.49 Light windows in dark mode, Git on Windows

Every instibrain window keeps dark text on a light background in dark mode, and on Windows the installer sets up Git for Windows before Claude Code. It also carries `instibrain mcp`, through which Claude Code on your computer reaches instibrain.

-   system

### How-to pages for 14 tools

The system holds a how-to page for each of 14 tools, the same for every customer, which your AI reads before it uses one: Codex, Chrome, Slack, Notion, Bitbucket, Semrush, Google Search Console, WordPress, Cloudflare, Google Cloud, Kindle, Spendee, TeuxDeux and WebWork Time Tracker.

-   sync client

### Claude Code on your computer works as you

Claude Code on your computer reaches instibrain through the sync client, signed with the computer's own keys, with no separate login. It reads and writes the spaces that computer syncs, at your role, and changes nobody's access.

-   cockpit

### Name your AI, and see whose each AI is

You rename an AI of yours in its row, with F2 or Rename. Every list of who reads a space or a specialist says whose each AI is, yours or a teammate's, in the cockpit and for your AI.

-   brain

### Confirmed changes land on the first try

A change your AI previews and confirms goes through whichever server answers the confirm. An AI that calls a tool by its old name is told the new name and to reconnect.

-   brain

### Each connected AI has a name and keeps its shares

When you connect an AI, the approval page asks which it is: a new one you name, or one of yours, with what that one reads. What you share with it outlasts each login, and removing it ends its logins and shares together.

-   sync client

###  0.0.48 A stalled folder retries every 30 seconds

When a folder's sync cannot start, your computer tries it again every 30 seconds instead of leaving the space unsynced until the sync client restarts.

-   cockpit

### Every connected AI and what it reads

The cockpit lists the account's connected AIs, each with whose it is. For your own, it shows what is shared with it and its logins with their last use, and you share with it, sign out a login, or remove it, with Undo.

-   sync client

###  0.0.47 Claude Code windows on Linux

`instibrain window` opens a Claude Code window on Linux, in GNOME Terminal, Konsole or xterm, with the window's name as its title and its own background colour.

-   system

### Google Apps Script: 83 actions, one authorisation

The system's Google Apps Script page lists all 83 actions with their parameters and traps. A Google Doc reads with every tab and its links, a replace hits one match, one authorisation covers every service, and an attachment comes back with no Drive copy.

-   brain

### Your AI reads only what you share with it

An AI you connect reads nothing until you share a space or a specialist with it, and never above your own role; the approval page says so. Its refusals name no space that is not shared with it.

-   cockpit

### The cockpit on a phone

On a phone, each row of a list stacks its columns as label and value, so nothing scrolls sideways, an empty value shows no label, and Home's six tiles take two rows.

-   cockpit

### List pages open on their table

Spaces, People, Computers, Specialists and Groups open straight on their table, with the counts in the filter pills instead of figure cards above the list. On a phone the first row sits right under the search.

-   cockpit

### Copy an invite link from the row's menu

An invited person's link is copied with Copy the invite link, first in their row's menu, where every action on a row lives.

-   cockpit

### A space's settings open as a sheet

A space's settings slide in from the right: its logo, what everyone it is shared with may read, who recovers its history, and Erase last, with its own step.

-   cockpit

### Share from the sheet's foot

Sharing a space or a specialist ends at one button in the sheet's foot: for a person or a group it first shows the pages they would read, then the same button shares.

-   cockpit

### A sealed space says what to do next

A sealed space on the Brain page shows its name, a Sealed badge, one sentence on what to do next, and the button to link a computer. Limit access on an AI opens Sharing with that AI chosen.

-   cockpit

### Rename a space, group or specialist in its row

You rename a space, a group or a specialist in its row, with F2 or Rename in its menu: Enter saves and Escape keeps the old name.

-   cockpit

### Linking a computer, step by step

The dialog that links a computer lists its steps in the order you take them, the installer first and then the code, and names the folder the space lands in: ~/instibrain/spaces/ and the space's name.

-   cockpit

### Wide tables stay inside the screen

A computer's row fits a laptop screen: up to three folders, each with its key's fingerprint shortened, the rest named on hover. A wide table scrolls inside its own box, and the page never runs wider than the screen.

-   cockpit

### A badge marks only the exception

Lists mark what differs and leave the normal state blank: a sealed space no computer has laid out shows Waiting for a computer, People marks who is Invited, and Sharing marks a group or an AI under its name.

-   cockpit

### Lists show 20 rows a page

Every list shows 20 rows a page until you pick another count, and the page controls appear only once the rows run past one page. A long description stays within two lines.

-   cockpit

### Screens state facts, not explanations

Sharing shows a space's facts as badges under its name: your role, Sealed, and how many computers hold its key or wait for it. The explanations above Spaces and Specialists are gone, and an empty Specialists page invites the first one.

-   cockpit

### A specialist opens on the Brain page

A specialist opens on the Brain page as a space does: its pages form the map, its hub opens first, and each page is read and changed in the same editor, with its versions and who wrote each.

-  26 October  2026
-   sync client

###  0.0.46 The desktop menu lines up every name

The desktop button's menu lines up every name, whatever its icon, and each session started from it takes its prompt colour.

-   brain

### Sharing has one set of words

Everyone a space or a specialist is shared with is 'shared with', and 'reader' names the read-only role alone. Your AI's tools follow: `share_space`, `unshare_space` and `shared_with`; a page's Held by tab lists the AIs and computers that hold it.

-   sync client

###  0.0.45 You approve each new computer of a sealed space

A new computer gets a sealed space's key only after a person on a manager's computer approves its fingerprint, in a dialog on a Mac or Windows or with `instibrain approve`. A rejected computer gets nothing, and nobody is asked again.

-   sync client

###  0.0.44 The sync client updates itself

`instibrain update` fetches the newest sync client, checks it as the installer does, and keeps your settings and links; the daily update runs it before Claude Code's own. `instibrain version` prints the version you run.

-   sync client

###  0.0.43 Sealed tool files go up once

A tool's file in a sealed space goes up once instead of on every cycle, and `instibrain sync` run beside the background sync says the background sync already keeps the folder, and ends cleanly.

-   sync client

###  0.0.42 Helper sessions run on Opus

The helpers a Claude Code session starts on your computer run on Opus, whatever model the session itself uses.

-   brain

### Your AI lists and unlinks your computers

Your AI lists your linked computers, with the spaces each one syncs and when it was linked and last used, and unlinks one after your yes. Nobody else's computers show.

-   sync client

###  0.0.41 Every change arrives once

A computer that catches up on a space receives each change once and in order, with none left behind, from its very first pull.

-   system

### Setup moves a tool's files into its space

When your AI sets up your computer, it shows you the files a tool kept in a hidden folder and moves them into the tool's folder in its space, where they run once a manager approves them.

-   system

### One set of rules for Claude Code windows

The rules for Claude Code windows hold for everyone: a window on a specialist takes its persona, one on an operation reads its folder, and the curator's window checks the space. A name you give picks the server, specialist, operation or project it points to.

-   sync client

###  0.0.40 One policy per computer, and changes in order

Your computer installs one policy into Claude Code: your Personal space's, else the first team space's, from its rules/policy/ folder, which only a manager changes. Each space's changes arrive in the order they were saved, never held back by another space.

-   system

### A tool has a system page and a space page

Each tool has two pages under one name: the system's says how the tool works for every customer, and your space's holds your account, ids and scripts. Your space's page only adds to the system's, so an improvement reaches everyone.

-   brain

### Your AI answers proposals

Your AI lists the proposals waiting for you, with who made each and who answered, and closes one as done or declined. The cockpit keeps Reject for a proposal you turn down.

-   cockpit

### An open tab survives an update

A cockpit tab left open through an update reloads itself once, instead of showing a dead page.

-   sync client

###  0.0.39 One folder per space when you link them all

`instibrain link --all` keeps each space in its one folder, so a computer linked long ago gets no second Personal or Sealed folder beside the first.

-   cockpit

### Proposals sit under Inbox

The inbox view lists proposals after its items, with the same rows and actions as Waiting for you, and the inbox count on the map includes them.

product  Sync client

What runs on your computer, how it updates itself, and every command.
product  Tools

Every tool the brain offers your AI, in the brain's own words.

---

instibrain is made by WebWork AI, Inc. Everyone uses the same models · the difference is what yours knows.
