Skip to content

Releases

rlx 0.10.1 — Delegate. Relax.

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 19 Nov 15:10
c6b523c

Introducing rlx: a manager for your AI coding agents, Claude Code and Codex. Tell the manager the goal. It plans the work, runs the coders, testers and reviewers, and keeps going until the goal is met. Then relax.

Install:

curl -fsSL https://aisquare.studio/rlx/install.sh | sh

On Windows, run irm https://aisquare.studio/rlx/install.ps1 | iex in PowerShell. The package on PyPI is aisquare-cli. Its command is rlx; relax, asq and aisquare start it too.

Upgrading from 0.10.0:

  • Run aisquare upgrade, or use the app's Update. Both add the rlx and relax commands to ~/.local/bin.
  • If ~/.local/bin already holds another program's rlx or relax, move it aside first: 0.10.0's upgrade replaces it without asking. Or run the one-line installer, which stops and names the file instead.

Added

  • Two new commands, rlx and relax, start rlx, as asq and aisquare do. All four run the same program: bare rlx opens the app, rlx --help names itself, and rlx --install-completion sets up completion for rlx. The one-line installer, 0.10.0's aisquare upgrade and the app's Update put the two new commands in ~/.local/bin beside the others. Hooks still call aisquare, whichever command connected them.
  • Another program's rlx or relax is never replaced or deleted. When ~/.local/bin already holds one (PyPI's relax is another project), the installer, rlx upgrade, the app's Update and rlx uninstall stop before they change anything and name the file: rename or move it, then run again. rlx doctor names one earlier on your PATH. Moving up from 0.10.0 with its aisquare upgrade, its Update or uv tool upgrade aisquare-cli replaces such a file without asking: move yours aside first, or use the one-line installer, which checks.
  • Pause sits beside Stop and Restart in an agent's view. A paused agent shows Resume in its place, which brings it back where it was, and its view says it is paused and what the pause keeps, where it read "(pane gone)" beside a pane id it no longer had. Pause, Stop and Restart each ask first in one small box, which says what happens and offers Don't ask again: tick it and that action acts on one click from then on, for every agent. Ask again before every action, under ⚙ Settings ▸ App settings ▸ User preferences ▸ Advanced, brings every box back. Resume never asks.
  • Each agent in the sidebar has a ⋯ menu: Pause or Resume, Stop, Restart and Rename. It shows on the agent you point at or have selected; a right-click or . opens it too. Rename is greyed, with why, where an agent keeps its name.
  • The Pause, Resume, Stop and Restart boxes for many agents list each one with a tick, and a Select all at the top, so you choose which ones they act on. Pausing only some of a project's agents leaves its manager free to start new ones; picking them all pauses the project as before.
  • App settings has a Display tab: on/off switches for the account usage after each agent, the agent marks in a mixed fleet, the live count in the sidebar's title, the fade of the side without the keyboard, the bell when an agent needs you, the board following its live feed, and the project directory under each name. Each applies as you tick it.

Changed

  • Personas live in Settings ▸ Personas, for you across all projects. A project's view has no Personas tab now, and the app offers your personas and the bundled ones.
  • Signing in to AISquare uses rlx's own page once the server offers it; until then, the existing page.
  • A manager waiting on you is not woken by noise. Watch's lines about its own results and questions, its ping for an agent idle by design, and a review that repeats a note it read no longer wake it. A manager that goes on standby (aisquare team standby), or asks you something with nothing in flight, is not woken by watch's routine lines until your next prompt.
  • You can see which pane has the keyboard. While the left pane has it, its highlighted project, agent or setting is a bar in the theme's colour; otherwise it is the dark highlight it always was.
  • The product is named rlx. The README is titled rlx and opens as rlx's page does: "Tell the manager the goal. Then relax.", what rlx does for you, and the one-line install. PyPI's summary and rlx --help say what rlx is in the page's words: "rlx is a manager for your AI coding agents. Tell it the goal: it runs Claude Code and Codex on accounts with room. Then relax." Five of the seven pages it links say rlx too. The package is still aisquare-cli on PyPI, and its keywords there name rlx.
  • The AISquare sign-in keeps its name: aisquare login, logout and whoami, aisquare init --api-key and aisquare explainability are AISquare's, and every page writes them so.
  • The sidebar no longer shows the AISquare sign-in; it lives in Accounts. AISquare's own messages say rlx where they said "the CLI": "the keys rlx minted", "Sign-in from rlx is temporarily paused".
  • The app's top-left shows rlx's logo, r\x, where it read Fleet. The live count beside it is as before (r\x · 3 live), and nothing else in the sidebar moves, at any width. The window's title and everything else still say rlx.
  • The repository is now AISquare-Studio/rlx, and every link says so: the one-line installers, the README and docs, rlx doctor's issue link and the plugins' manifests. Links to the old name keep working, since GitHub redirects them, and the package, the plugin and its marketplace keep their names.
  • Usage sharing, once you turn it on, posts to /api/v1/rlx/metrics/, and sends the same counts as before. Releases up to 0.10.0 post to /api/v1/cli/metrics/, which AISquare keeps answering until May 2027. A server that answers 503 with Retry-After is asked again after that wait, as a 429 already was, not after the first five-minute retry; either waits a day at most.
  • The one-line install is shorter: curl -fsSL https://aisquare.studio/rlx/install.sh | sh, and on Windows irm https://aisquare.studio/rlx/install.ps1 | iex. Both lead to the same installers on GitHub, and a release's own pinned URL is unchanged.
  • rlx names itself everywhere it speaks to you. Its messages, errors, --help, the doctor and the app's screens say rlx where they said aisquare, asq or "the CLI": "Quit the other rlx?", "rlx 0.10.1 is up to date", Update rlx, the window title. rlx --version prints rlx 0.10.1; installers of earlier releases read the version from it as before. No --json key changes, and no machine value (names, statuses, ids), the doctor's asq instances row included; only the human text inside a value says rlx.
  • Every command it tells you to run starts with rlx: rlx accounts add, rlx doctor, the doctor's one-click fixes. When another program's rlx comes first on your PATH, those lines say aisquare … instead, so a line you paste always runs rlx. Your agents' instructions keep aisquare …, the same commands.
  • The installers say rlx too, and end with "Open rlx with: rlx". The Remote page is titled rlx remote; a phone that added it to its Home Screen shows the new name once it is added again. The plugins describe themselves as rlx's; their install names (aisquare@aisquare-cli) do not change.
  • The package's page on PyPI links to rlx's site and its docs, and the README links the site.
  • The board names each agent as the left pane does, where it showed role·<id>. In the app's board, board -w and rlx board, a line's author, the sessions block, a task's claim and the done archive read the agent's name, such as coder-auth, or its role where the left pane has none. A note you wrote from the command line reads you, and owner in aisquare board, which agents run (it read cli). An agent keeps its name after it stops, is cleared or moves to another account, and a renamed one shows its new name. Of two sessions with one name, the earlier adds the time it started: coder-auth (14:02). Long names end in …, the short id stays in a line's detail, and o still opens the right transcript. Agents still see ids, and every --json is unchanged.
  • The app opens in rlx, a theme of its own, until you pick another. It is built on rlx's blue, #2F74F0, with teal, lime, green, amber and coral on near-black. Press t to pick another theme, as before. A theme you picked earlier is kept, and so is one you set in TEXTUAL_THEME. Every theme it offered is still there. rlx board -w shares your pick, so it opens in rlx too.
  • The logo at the top-left, r\x, and Welcome's heading take their colour from the theme. Under rlx they are #2F74F0. Under any other theme they take its main colour, or its text colour where the main colour is too faint to read as text (textual-dark's is), so they change when you press t. They keep that colour while the other pane has the keyboard. Only colours change: nothing moves, at any width. rlx cannot set your terminal's font. JetBrains Mono, the font on rlx's web page, is the one we recommend.
  • The app never asks you to type a command. Its screens, notifications, tooltips and dialogs point to a button or a page instead: a missing agent or tmux links its install page, an install rlx cannot update says to update it the way you installed it, and the Doctor's buttons say what they do ("Connect Claude Code"). A check that needs the command line offers Ask the manager, or says a manager can fix it. The messages it shares with the command line leave the command out in the app, the usage-limit and pause lines included; the agents still read it. The command line keeps every command, byte for byte, and so does everything an agent reads.
  • Notifications are small, sit top-right and are quick to clear. They start under the top row, clear of the agent bar's buttons, at most 46 columns wide. The newest is on top and at most three show; the rest wait, their time stopped, under a `+N m...
Read more

0.10.0 — Claude Code + Codex support, Codex in beta

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 16 Nov 17:20
47f1444

Codex support, in beta: run OpenAI's Codex CLI beside Claude Code, or instead of it, with
the same memory, board, fleet, accounts and asq, and hand work from one to the other. Codex
is marked beta wherever you choose or see an agent, until a release has run without
complaints. Claude Code now runs through the same adapter as Codex and does what it did in
0.9.2; what changes for every user is listed below.

Upgrading:

  • Run aisquare upgrade, or use asq's own Update button. From a shell, quit asq first
    (and aisquare serve or remote serve): one left open runs 0.9.2 until it is started
    again.
  • A 0.9.2 installed fresh since 2026-11-09 fails on every command, aisquare --version
    included (typer 0.29.0, under Fixed), so it cannot run aisquare upgrade: run the
    one-line installer again, which repairs it, or uv tool upgrade aisquare-cli. A 0.9.2
    that works upgrades with aisquare upgrade, as above.
  • aisquare upgrade keeps your config.toml, hooks, accounts, their order and the
    default (rank 1). The store goes from 0.9.2's v43 to v50 on the first command after the
    upgrade, in about 0.3 s on a store of 10,000 board events. Each step adds a table or a
    column and nothing is removed: the rows you have are marked as Claude Code's, and your
    account order becomes the one order below, so nothing moves.
  • Agents already running keep working: their hooks call 0.10.0, and fleet ls, tell,
    stop, restart, switch and the usage-limit hand-over reach them, and any agent a
    still-open 0.9.2 asq starts. A restart or switch resumes the agent's session, as 0.9.2's
    did.
  • Going back with aisquare upgrade --version 0.9.2 works on the v50 store: 0.9.2 opens
    it as it is, and everything 0.10.0 added stays for when you upgrade again. While on
    0.9.2, an account reorder or a new default is dropped when you upgrade again, and an
    account you add is ranked last. Agents 0.9.2 spawns, restarts or switches meanwhile are
    managed by 0.10.0 when you upgrade again. aisquare upgrade --version 0.9.2 says what
    it keeps and loses before it moves back. A Codex account's hooks are left as they are on
    the way back, since 0.9.2 runs none; the move names the
    aisquare agents disconnect codex --config-dir … to run first. A hook the move back
    could not rewrite is a warning with the command that rewrites it: the move back no
    longer ends with an error for its hooks.
  • Codex accounts across a rollback: 0.9.2 numbers its accounts by Claude Code's alone, so
    an account you add there can take the number of a Codex account a pre-release 0.10.0
    build made. 0.10.0 lists the newer of the two last. The Codex account keeps its place,
    and the newer one is never moved or made the default until it has a number of its own;
    aisquare doctor names the commands that give it one. Wherever you name that number
    (--account 5, accounts remove 5, a role bound to 5 with no agent), aisquare names
    both accounts and acts on neither: name one as codex:5 or claude-code:5, which every
    command that takes an account accepts. To bind a role to one,
    aisquare team bind <role> --agent claude-code --account 5 is the form 0.9.2 still
    reads after a rollback, and a role bound to an agent reads 5 as that agent's.
    team bind takes a change only where the role's launch would run it, from any folder:
    so --agent claude-code with a Codex account's number, or a later --agent or --bin
    on a role whose bound account is another agent's, is refused, with the reason every
    launch of that binding would give. A bind of --env or --arg alone is never refused
    for the role's account, as on 0.9.2, even while that account is removed or signed out.
    aisquare accounts list shows a shared number as codex:5 or claude-code:5; a
    number one account holds reads as before. A Codex account keeps its number from 0.9.2
    once 0.10.0 has read the accounts (any aisquare accounts command, a launch, asq) or
    moved back to 0.9.2. Upgrading to 0.10.0 keeps it too, where aisquare's hooks are
    connected.

Codex support (beta)

  • How to start. Install Codex CLI 0.164.0 or newer (install.sh --agent codex or
    --agent both installs a missing one), then run aisquare agents connect codex. Pick
    Codex (beta) in the Spawn dialog's Agent field, or run aisquare fleet spawn <role> --agent codex; aisquare team bind <role> --agent codex makes it a role's agent.
  • What works. Memory, the board, the fleet (spawn, restart, switch, tell, stop, pause
    and resume, fleet top, fleet adopt), Welcome, doctor, Usage & stats, personas and
    Remote, each through Codex's own hooks, flags and screens. Teams can mix agents, and a
    restart resumes the agent's Codex thread.
  • Which agent is which. In a fleet of more than one agent's, fleet ls names each
    row's agent after its role (Claude Code, Codex (beta)), and asq's sidebar rows carry
    the agent's mark after the role icon (✻, >_). With accounts of more than one agent,
    as the Accounts page marks its rows, asq's Agent view header and Manager tab name the
    agent after its account and use (codex-1 · 42% codex), and team harness says each
    role's after its model (agent=codex [role]); the account picker (Spawn's Pick…)
    names each account's agent after its number, the agent's mark first
    (5 · >_ Codex · …); Adopt a conversation… names each conversation's agent when it
    lists two agents', as fleet adopt --list does; and asq's New bind form has an
    Agent field, as team bind --agent does, also on a binding that runs Codex.
    aisquare accounts usage names a Codex account's agent after its name (account 4 · codex, plain codex · codex), before its first session as well as after. The Manager
    tab tells a Codex manager to type to it "as you would to any Codex session", and a Codex
    manager's cycle says "an agent that hits its Codex usage limit". An agent resumed after
    fleet switch or the hand-over is told it moved to "another Codex account". Adding
    aisquare's own home as a project warns that it holds "every Claude and Codex account's
    login data" once Codex has accounts here. Otherwise these read as in 0.9.2.
  • Accounts. aisquare accounts add --agent codex adds a slot and signs it in with a
    device code (codex login --device-auth): Codex prints a link and a one-time code, and
    you open the link on any device, phone included, and enter the code. Nothing waits for a
    browser on this machine, so it works over SSH and on a machine with no screen. + Add
    and Log in in asq's Accounts page sign in the same way. Device-code sign-in must be
    turned on for your ChatGPT login: in ChatGPT's security settings, or for a workspace, by
    its admin in the workspace permissions. Where it is off, aisquare accounts add --agent codex --browser signs in with the browser as before (codex login); a sign-in that
    ends with no login says both. Usage is read off each account's own sessions, and an
    agent at its limit can move to the Codex account with the most room. Until a Codex
    session has recorded a reading on an account, asq's Accounts page reads no reading yet
    there, dim, with the reason on hover, not the ! of a broken login.
  • Your own Codex home (~/.codex, or $CODEX_HOME) is one of your accounts only once
    aisquare has taken it on: connected (agents connect codex, install.sh --agent codex),
    an account of Codex added (accounts add --agent codex), signed in from Welcome, or
    arranged by its number (accounts alias|default|move|order|disable 0). Until then it
    takes no launch that nothing pins and no hand-over crosses to it, so a Claude Code user
    with Codex merely installed and logged in launches and waits for a limit's reset as on
    0.9.2. --account 0 still names it.
  • Doctor grades the Codex home your shell names (CODEX_HOME) only when Codex is your
    selected agent, and a Codex home only aisquare's records name (0.7.0's agents connect codex) only where Codex is installed: with no Codex here, its row reads ok, as on 0.9.2.
  • Hand-off across agents. fleet switch --agent <agent>, or Hand off from in the
    Spawn dialog, starts the other agent fresh with a brief of the work. The brief goes to
    the other agent's vendor, with credentials scrubbed out first.
  • Sandboxed. A Codex agent's sandbox writes nothing in the store: its board calls go
    through a broker that runs outside it.
  • Reviewer denials. Under --approve-for-me, a command Codex's automatic reviewer
    denies counts as a denied permission in aisquare outcomes, as one you refuse at the
    prompt does. That includes a command inside a code-mode script, how Codex runs every
    call on GPT-6.1-Sol.
  • Telling a Codex agent. Codex reports a turn over a moment before it accepts
    commands, so a fleet tell right after a Codex turn waits for that moment (about a
    second, at most 5 s): a told /clear or /compact is not refused "while a task is in
    progress". Plain text is never lost: Codex holds it and runs it after the turn.
  • -a is aisquare's. In launch and fleet spawn, -a N is the account
    (--account) for a Codex role too, and -c BIN (--command BIN) a program run as the
    role's agent (Codex's -c KEY=VALUE still reaches Codex). Give Codex's own -a as
    --ask-for-approval or after --, or set it with fleet spawn --approval;
    fleet spawn … -a never is refused with a line that says so.
  • Known limits.
    • Codex runs a hook only once it is trusted. agents connect, accounts add and
      aisquare upgrade record that trust for you; where they cannot (a config.toml
      aisquare may not write, or the Codex plugin's hooks), trust aisquare's hooks in
      Codex's /hooks. Until then doctor says so, and the board sees nothing from that Codex.
    • CI tests Codex 0.164.0, 0.166.0 and 0.167.0, all pinned, 0.167.0 on macOS too, where
      the tests o...
Read more

0.9.2 — multi-instance asq, plus fixes and optimizations across the fleet

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 03 Nov 21:06
6a77e65

Run asq at your Mac and over SSH into the same Mac at once, on one home and one fleet: each asq shows the others, the one over SSH shows the Mac's usage, and an asq of another release won't start beside a running one. Around that come fixes and optimizations across the fleet. Pause a project's agents, or every project's, so idle agents stop spending tokens, and resume each later on its own session, claim and worktree, even after a reboot. Spawns follow the machine's load instead of a fixed four per project. asq asks fewer questions and asks them in place, buttons work as soon as you see them, and a double-click never confirms the dialog it opens. aisquare outcomes says what your agents got done this week, and aisquare fleet top what each one consumes. An idle account whose token went stale is refreshed quietly instead of reading as signed out. Under the hood, tests that failed on Windows or now and then are fixed, and Dependabot's four version bumps are in.

0.9.2 builds on 0.9.1. It gathers these pull requests:

  • #472: several asqs on one home. It combines six pieces: #461 (usage over SSH), #462 (the instance gate), #463 (the pane size), #464 (the instance view and the SSH tmux guard), #465 (test dates) and #466 (the look for an older asq in ps).
  • #457: pause and resume.
  • #485: five items, each from its own branch: the quiet refresh of a stale login, real repository names and the sidebar's live count, aisquare outcomes, auto admission, and aisquare fleet top.
  • #491: the second release review's fixes in #485's code: auto admission never holds a manager or a resume, and max_agents_per_project takes auto or a number of at least 1.
  • #489: fewer clicks, from three pull requests: Stop's Don't ask again (#477), yes/no questions asked in place (#486) and a click guard that waits only until the screen has redrawn (#478).
  • #495: the third release review's fixes: a double-click never confirms the dialog its first click opened, Unset in a role binding's Unset key… dialog is guarded too, and a test flake.
  • #455: rename an agent.
  • #456: a project's rename reaches its codename and tmux session.
  • #458: one sidebar highlight.
  • #475: three asq features, each from its own pull request: the faded side (#470), the active-only filter (#467) and the Onboard view's ↻ (#468).
  • #454: Usage & stats and the Accounts table.
  • #469: TahirRabia's fixes, #447 to #450.
  • #479: the opt-out notice.
  • #494: CodeQL alert #23 in aisquare outcomes.
  • #488: Dependabot's #481 to #484 (setup-uv, certifi, ruff and truststore), with a Windows test fix.
  • #459: a flaky test.
  • #476 and #480: Windows CI for #472's tests, and the review fixes for #469 and #472.

#455, #456 and #457 each went through several rounds of review and testing. #457 was also run against 0.9.1's own code on a store it had paused, and a UI tester checked #454 on screen at 160x48. Each of #461 to #466 was tested and reviewed on its own before #472 combined them. #472's SSH guard rests on a probe from a real SSH login into macOS 27.2. #469 brought in TahirRabia's four fixes after three review rounds and a tester's round. Each of #485's five items was tested on its own, and #485 was approved in its fifth review round. #467, #468 and #470 were each tested on their own and validated together before #475 combined them. #477, #478 and #486 were each reviewed and tested by hand in a real terminal, with real mouse clicks, before #489 combined them. #495's double-click tests fail without its fix and pass with it. #488 and #495 each ran the full CI matrix, Windows included, before they merged. The release ran the full CI matrix green: the test suite on Linux (Python 3.11 to 3.13, with the captain on and off) and on Windows in 10 shards, and the installer checks on Linux, Windows and macOS.

asq at the Mac and over SSH, at once

You can now run asq at your Mac and over SSH into that same Mac at the same time, on one home and one fleet.

  • Each asq sees the others. While two or more are open, a line under the sidebar's status line says how many there are and where each runs: 2 asq · Mac + ssh (here). Hover it to see each one's version, host, pid and terminal. aisquare instances lists them (--json too), and the doctor's asq instances row warns when one runs another release.
  • One release per home. An asq won't start beside one of another release. It names the other asq and asks Quit the other asq? [y/N]. On y, the other quits as if you had pressed q in it, and its terminal says which asq asked. On n, the new one starts nothing and tells you how to quit the other yourself. It checks before it opens the store or touches tmux, so an asq that is refused has changed nothing.
  • A 0.9.1 asq is found too, and never killed. 0.9.1 writes no record, so the new asq looks for one in ps. It names that asq's pid, terminal, start time and command, and says what kill <pid> would lose compared with quitting it with q.
  • Usage over SSH shows what the Mac recorded. An SSH login can't read the login Keychain, so in 0.9.1 its asq showed no usage. Now it shows the readings the Mac's asq took for the same login, never another login's, marked · Mac and via the Mac's asq · 2m ago. When the Mac sees a sign-out or an expired login, the SSH side shows it too.
  • A shared pane takes the size of the asq you last typed in. When both asqs show one agent, a key, a paste or a click makes that asq size the pane. The other asq shows [sized by the other asq (ssh)] instead of resizing it back. In 0.9.1 the two kept undoing each other's size.
  • SSH doesn't start the fleet's tmux server on macOS. Agents in a server started from an SSH login can't read the Keychain, so a start from SSH is refused with start asq on the Mac first. Once the Mac has started the server, agents spawn into it from SSH as usual.
  • Upgrade and uninstall wait for the other asqs. aisquare upgrade and aisquare uninstall refuse while an asq of 0.9.2 or later runs on the home, and asq's own Update and Uninstall refuse while another one runs. Each refusal names the asqs to quit first.

Running more than one asq has the details.

Pause the fleet, and pick up where it stopped

You can now stop your agents spending tokens without losing their place, and bring them back when you are ready.

  • Pause one project, a group or all of them. aisquare fleet pause exits each running agent's Claude process as fleet stop does, the manager first, and keeps its label, task claim, worktree, branch, account and transcript. -P <project>, --group <group> and the new --all and --managers choose which agents. The captain, and an agent that runs the command, keep running.
  • Resume brings back exactly the agents a pause stopped. aisquare fleet resume starts each one under its own label, resumes its session from the transcript (claude --resume) and types one carry-on line, with the manager last. An agent that exited without a pause stays 💤, and running resume twice does nothing the second time.
  • Nothing is handed out while they are paused. A paused agent keeps its claims however long the pause lasts. The board says agent_paused, which wakes nobody, instead of agent_exited.
  • Pause before a reboot or an OS update. A paused agent outlives its tmux server: fleet reap --server-down leaves it paused with its worktree and claims intact. reap, Restart all, Stop all and fleet shutdown leave it alone, and a new spawn takes the next label instead of its label.
  • Pause and resume from asq. ⚙ Settings ▸ Fleet has Pause all… and Resume all… beside Restart all… and Stop all…. A project card's ⋯ menu and its Settings tab have Pause agents… and Resume agents…. Each opens one dialog that lists, by project, the agents it will act on. A paused agent shows ⏯ paused in the sidebar and in fleet ls. When a project's manager is paused, its Manager tab offers ▶ Resume agents….
  • Resume one agent, or the whole project. A paused agent's view offers Resume and Stop. Resume brings back that agent alone, as fleet restart <label> does, and the project stays fleet-paused until Resume agents… or fleet resume. Stop ends the pause for good.
  • Only you end a pause. You end one from asq, Remote or your own shell, with no --as. When an agent tries to stop, restart, switch or resume a paused teammate, the command is refused whatever --as it passes, and the captain acts as you only from its own window.
  • A paused agent whose task closed is not woken for nothing. Your resume ends its pause instead, returns its claims to the board and says so. A paused manager whose task closed stays paused until your fleet stop manager.

Spawns that follow the machine's load

By default, 0.9.2 no longer stops each project at four agents. It decides by how busy the machine is.

  • max_agents_per_project now defaults to auto. A spawn is admitted unless the machine is busy. While it is busy, a spawn into a project that already runs 3 live agents is held, and a project with fewer is always admitted.
  • Busy means any one of these: a 5-minute load average of 1.0 per logical CPU; on Linux, CPU pressure (some avg60) at 40; memory pressure at warn; on macOS, thermal pressure at heavy. Each threshold, the floor of 3 and the hold are [fleet] busy_* settings.
  • A fan-out is paced even on a calm machine. The load average moves minutes after a burst of spawns does, so past the floor a project gets at most 2 new agents every 5 minutes. Ten spawns started at once into an empty project leave 5 running, and the rest are refused as held.
  • A held spawn never waits. It is refused at once with capacity_busy, and the message names the signal, its value, its threshold and the earliest retry. While aisquare watch runs, it wakes the project's manager when ...
Read more

0.9.1 — see what you spend, run the model you chose, and start fresh

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 27 Oct 19:08
d2f8d3f

See where your tokens go: Usage & stats breaks down every Claude Code session on your machine by day, repository, account and model, so you know what is spending your allowance. Spawns run the model and effort you picked, so a manager no longer starts on Fable at high effort unless you chose it. Reset a project's board and fleet with one command, aisquare project fresh, and the next task starts clean. And more: your fleet's headroom as one number, a dropped screenshot that reaches the agent on screen, a conversation you started by hand that can join the fleet, and solo MCP sessions that stay solo.

0.9.1 builds on 0.9.0 and gathers #432, #397, #442, #436, #429, #298, #441, #444, #446 and #451, with the sharded CI of #427 and ruff 0.17.2 (#380) underneath. #397, #429, #441, #444 and #298 each went through several rounds of review. The release ran the full CI matrix green: the test suite on Linux (Python 3.11 to 3.13) and on Windows in 10 shards, and the installer checks on Linux, Windows and macOS.

See where your usage goes

You can now see how much room your accounts have left, and where your tokens went, without leaving asq or Claude Code.

  • One number for the fleet's headroom. The status line under ⚙ Settings shows the share of your accounts' allowance left right now, ● 3× claude 58%, and the Accounts page and aisquare accounts usage end with a one-line fleet summary.
  • The manager's team, in one line. A strip under the Manager tab counts the agents it runs by role and state; click it to open the Board.
  • Your usage inside Claude Code's own status line (opt-in). aisquare accounts statusline --install shows the account, its 5-hour and weekly use, the fleet's share left and context use in every session, hand-launched ones too. --wrap keeps your own status line, and --uninstall gives your file back. It keeps a short record of each session for a day (never a prompt), which capture.enabled = false does not stop and --uninstall does.
  • A page for what you spent. ⚙ Settings ▸ Usage & stats shows tokens, cost at list prices, sessions and turns for Today, 7 days, 30 days or Lifetime, with your biggest repositories, accounts, models and busiest days; aisquare metrics stats prints the same. Every session counts, even ones aisquare did not launch, and it all stays on this machine.
  • It counts every repository you run Claude Code in, including ones you have not added to aisquare, whatever [projects] capture says. project forget --purge now makes a forgotten project stay forgotten.
  • Share anonymous daily counts, only if you choose to. It is off by default; asq asks once, and aisquare metrics share on, off, status and preview control it, with preview printing the body the next shipment would post (a labelled sample while sharing is off).
  • What goes, and what never does. Daily counts of tokens, sessions, commands and pages, plus a salted hash and four coarse traits per repository; never names, paths, remotes, prompts, emails, aliases or command arguments. If you are signed in to AISquare your counts are linked to you, DO_NOT_TRACK=1, AISQUARE_NO_TELEMETRY=1 and any CI run force sharing off, and docs/usage-telemetry.md has the detail.
  • Move an agent before its account runs out (opt-in, off by default). Set [accounts] before_limit = "switch" and an agent whose account reads at or over switch_at (85 by default) moves to another account at its next turn end, never mid-turn; "warn" only writes a line on the board.
  • An account with no usage to show reads !, and says why on hover. Hover for the cause and the fix, such as a locked login Keychain; accounts usage prints the same.

Spawns run the model you chose

You choose a model and effort once, and fleet spawns run them.

  • A spawn starts on what you chose, not the ladder's top rung. 0.9.0 started every manager on --model fable --effort high over your saved default; now Fable or max run only when you chose them, and with no choice Claude Code runs the account's own default.
  • /model and /effort in any fleet agent are remembered. Your last pick is kept per aisquare home for every role and persona, and a restart or an account switch keeps what that agent was running.
  • Override one spawn without moving your pick. A spawn's own --model and --effort (or the Spawn dialog's new Model and Effort fields under Advanced) win over a role's settings, which win over your last pick; aisquare fleet restart <label> --model … --effort … runs one agent on them.

Start fresh, and a tidier fleet view

You can reset a project you reuse in one command, and spawn from one place in the sidebar.

  • Reset a project without onboarding it again. aisquare project fresh (alias project reset) stops its agents, removes their worktrees, clears its board and work briefs and starts a blank manager, so the next task is #1 again. Your checkout, other branches, settings and every other project are never touched, and project memory, the prompt log and fleet branches stay unless you add --memory, --prompts or --branches.
  • It will not quietly throw work away. A worktree with uncommitted or unpushed work stops the reset unless you pass --discard-unpushed, and --dry-run prints the plan and changes nothing.
  • In asq, one dialog shows the counts. ↺ Start fresh… is on a project card's menu and on its Settings tab, and unpushed work gets a second dialog before anything is discarded.
  • One Start captain and one Spawn agent in the sidebar. Spawn agent opens the Spawn dialog with a new Project field, and the Account field now says what a spawn with no account runs on, such as (default: rank 1 · work).
  • A sign-in brings you back to where you started it. Welcome, the persona picker and the Spawn dialog come back once a new account signs in, instead of leaving you on the Accounts page.
  • The Board says what needs you. A top line counts waiting agents, questions, reviews and blocked work, each a button to what it counts.
  • Role bindings in App settings. Agent settings ▸ Advanced lists every role aisquare team bind bound, to edit or clear, and aisquare doctor warns about a binding whose directory has no aisquare hooks.

Bring more into your fleet

You can hand the fleet a screenshot, a conversation you started by hand, or a session you continue from claude.ai/code.

  • A screenshot dropped on asq reaches the agent on screen. A paste no focused widget takes now goes to the one live pane (with two on screen, you click the one you mean), and under WSL a lone Windows path becomes its Linux path.
  • Adopt a conversation you started by hand. aisquare fleet adopt <session-id|transcript> resumes it as a fleet agent on its own account and in its own directory, keeping its session id; aisquare fleet adopt --list shows what is free to adopt, and the Accounts page has Adopt a conversation….
  • Claude Code's Remote Control, per role (off by default). Set [fleet.roles.<role>] remote_control = true, or pass fleet spawn --remote-control for one agent, and you can continue that session from claude.ai/code or the Claude app. Anyone signed in to that Claude account can drive such an agent.
  • Edit a persona in your own editor. In the Personas tab, Edit opens $VISUAL, $EDITOR or vi in a pane in place; Windows, or a machine without tmux, keeps the text box.
  • Import several personas at once. aisquare persona import takes several sources or --from-dir DIR and skips a README or LICENSE beside your skills, which cut the sample library's engine calls from 4 to 1.

Solo sessions stay solo

You can use aisquare's board tools over MCP in a solo session without it being treated as a team.

  • A solo session no longer carries about 1.1k extra tokens a request. In 0.9.0 its own MCP client counted as a teammate, so for 30 minutes after a board-tool call it got the board and the full team context.
  • Its own aisquare serve --stdio now proves it belongs to the session and acts as the session, however you mount the server, so the session's notes and claims show as its own.

The captain's owner gate

With the experimental captain on, the tools that act as you now need a code from your authenticator app.

  • The tools that act as you need a code. press, paste, stop, spawn, restart, wololo, attach_persona and act refuse outside a 15-minute window that a six-digit code opens, spoken to the captain or typed with aisquare captain gate unlock <code>. Reads, tell, notes, tasks and the bt brake never need one.
  • Set it up once. aisquare captain gate enroll prints a QR for your authenticator app, and it needs the new gate extra, pip install 'aisquare-cli[gate]'; repeated wrong codes lock entry, from five minutes up to 24 hours.
  • A spoken code can be overheard, so you can turn the gate off: add gate = false under [captain] in ~/.aisquare/config.toml by hand, since aisquare config set does not take it. The captain stays experimental, and off until you turn it on.

Less to carry

You get a simpler aisquare, with less running in the background.

  • The project brain (gbrain) is gone, and so is the background drain every board write started. aisquare recall, aisquare team distill, the recall MCP tool (eight tools now), doctor's brain row and the AISQUARE_BRAIN* variables are removed. Nothing on disk is deleted.
  • Explainability and the CI test bed are parked. They are out of aisquare doctor and every screen, and still run through their own commands, such as aisquare explainability doctor.
  • The docs/ directory is gone from the repo, except the page on what usage sharing sends. Run aisquare --help for the commands; 0.9.0's pages stay readable at the v0.9.0 tag.

...

Read more

0.9.0 — the captain, personas, the evidence gate, and one-click updates

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 26 Oct 02:12
02e99bf

The captain: an experimental voice-and-text agent that runs every project's fleet for you. It is off until you turn it on. Personas let you run any agent as someone. Remote control needs nothing more installed. The fleet admits spawns against the machine's memory, finished workers stop by themselves, and a symbol index (aisquare code) answers where code is defined and used. Work briefs and the evidence gate: a task linked to a brief's requirements is done only on a fresh pass from an asq exec report. Every first message reaches its agent, nothing the fleet types lands in your draft, agents carry far fewer tokens, restart all brings back agents that died, the agent pane keeps up with your typing, the first enabled, signed-in account is the machine default, asq offers each new release and installs it in one click with your fleet still running, every release now carries its own installer, and aisquare no longer packs your codebase or needs Node.

0.9.0 gathers everything since 0.8.2: #201 (personas, the hackathon release train), #240 (the captain), #267 (the agent info line), #272 (tmux control mode), #273 (certifi), and #274. #274 is the follow-ups, with #302 (App settings in tabs), #303 (the simple Spawn dialog) and #346 (the solo-session trim and token metrics) folded in. Its closing reviews came back clean, with no blocker or major, at 0545dc5 and again at 189dffd. Its last gate ran the full suite in an isolated home with the captain on (7439 passed, 0 failed) and off (7244 passed, 0 failed). It merged with all 24 CI checks green, windows-latest included. The release PR adds the versioned installers and the version bump, and folds in six late changes: #405 (the Spawn dialog's follow-ups), #406 (rank 1 is the machine default), #301 (the daily release check and one-click update), #408 (usage reads back off on HTTP 429), #403 (every HTTPS call verifies against the OS trust store) and #402 (the one-liner on main runs the release's own installer). It also brings in main's P0 reliability train (#295, with #409): first messages, typed writes, idle agents and account switching; and its P1 token diet (#296): leaner agents, a token ledger and a quieter manager. It also folds in #399, which removes the codebase snapshot and Repomix with it, #417 (the evidence-gate stack: command reports, work briefs and the gate on task done), and main's P2 train (#297): a symbol index, spawns admitted against the machine's memory, finished workers that stop themselves, one watcher with no model, parent managers and shared account memory.

The captain (experimental)

  • One agent over every project's fleet, by voice or text. It is one Claude Code session per home. Its only tool is its own Actions server, which has 24 fixed tools, and every call is audited with your words and its receipt.
  • It unblocks agents for you. It answers an agent's Claude Code permission prompt (press yes, press no), types into an agent's pane, and spawns, stops and restarts agents. A stop, spawn or restart acts only when your words name the agent, its role or its project. Otherwise the captain asks one question first ("Stop coder-1 in alpha?"). It never types into an agent's trust dialog.
  • The attention queue folds every board into one ranked list of what needs you: questions, blocked or waiting agents, reviews, pull requests and stale work. See aisquare captain attention | next | resolve | snooze.
  • The voice page (aisquare captain --voice, the [voice] extra). Hold to talk, or leave it listening for the wake word; only "Captain, …" gets through. The reply is spoken back.
  • It ships off. Turn it on in ⚙ Settings ▸ Captain, which applies at once with no restart, or with aisquare config set experimental.captain true, which a running asq reads at its next start. While it is off, none of it runs; asq shows only its switch, and doctor a captain: off row. Guide: docs/captain.md.

Personas

  • Run an agent as someone, such as a skeptic, a mentor or a minimalist. Use aisquare launch <role> --persona NAME, fleet spawn --persona, or a default per role ([fleet.roles.<role>].persona).
  • A persona is a Claude Code skill. aisquare persona list | show | new | edit | rm | validate | import | export manages them in three layers: the project, the user, and four bundled personas. A skill copies in and out byte for byte.
  • Import anything. Plain text, another tool's persona, or a web page is converted with an LLM, shown to you, and confirmed before it is saved.
  • Attach to a running agent with aisquare persona attach <name> --to <label>. A restart keeps the agent's persona.
  • In asq: a Personas tab in the Project view has a live preview, an in-place editor, import, export and a two-step attach. The Spawn dialog has a Persona field, under Advanced.

The fleet in asq

  • The simple Spawn dialog. It shows three fields: Role, Hand off from, and Account. Everything else folds under Advanced. There is also a new role-less option, (none) — general-purpose Claude, which is aisquare fleet spawn agent or aisquare launch agent from a shell: a plain Claude Code session on the board, with no standing cycle.
  • App settings in tabs: Agent settings, User preferences and Remote. The Captain box sits at the top of Agent settings. Only the essentials show; the rest folds under Advanced. The first preference is Always show project directory.
  • Stop an agent from its view's header, or with x in the sidebar.
  • Rank 1 is the machine default. There is no separate default account any more. When nothing more specific picks one (no --account, role binding, project default or headroom pick), a launch runs on the first enabled, signed-in account in the priority order, and its line reads [work · preferred (rank 1)]. aisquare accounts default <ref>, with no --project or --role, now moves that account to rank 1, and a machine-level accounts default --clear is refused, since some account is always first. Project defaults and role bindings are unchanged.
  • A simpler Accounts table. The ★ column, the Default button and the status column are gone. ↑/↓ are the first columns of each row and the plan is the last. Why a row has no usage (not signed in, a token to renew, a usage read that failed) reads in its session and week cells, with the whole text in their tooltips. accounts list and accounts usage drop the * mark (the preferred account is the first enabled, signed-in row), and the account JSON drops is_default.
  • Usage reads back off when they are rate-limited. When the usage endpoint answers HTTP 429 (or 503 with Retry-After), the account's reads wait: the endpoint's Retry-After, or one minute doubling per 429 in a row, up to 30 minutes. The wait is recorded on disk, so no aisquare process, the Accounts page, --refresh, a headroom pick or a hand-over asks that account again before it ends. The wait is the signed-in account's: a token refresh or Log in again keeps it. Before, every poller kept the account rate-limited.
  • The last good reading stays. A read that fails keeps the most recent numbers, with their age and why the newer try read nothing: 40% · 12m in the page's session cell with the reason in its tooltip, and account 2: rate-limited until 14:32 — …; showing the reading from 12m ago under accounts usage and accounts list --usage. — shows only when nothing was ever read. --json readings carry kept, rate_limited, rate_limited_until and last_failed_reason.
  • A kept reading is never taken for a live one. A headroom pick uses it only inside the pick's own window with neither window reset since, and otherwise skips the account with a note giving its age. A hand-over whose live read fails with anything but a 429 or a login problem picks by a reading at most five minutes old, and says so.
  • Log in again on any signed-in row of the Accounts page, so a fresh login needs no aisquare accounts run <slot> and /login by hand. It does not clear a rate limit, as its tooltip says. aisquare doctor lists accounts whose reads are backed off, and until when (claude-account-backoff).
  • The agent info line. Each agent has one line showing its label, state, account with the share of its five-hour session used, model, effort, pane and persona. The sidebar rows show the session use too.
  • Two things moved. The agent view's header now shows that info line instead of the role, the task and the working directory. The command that attaches a terminal to a project's tmux session moved from the Manager tab to the project's Settings tab.
  • Faster panes. A key now reaches the screen in about a millisecond; before, it took 60 ms or more. Scrolling follows your gesture while you make it. The visible pane talks to tmux through one control-mode client per server; on tmux older than 3.2, the pane polls as before.
  • No more stall every two seconds. The sidebar refresh now reads in the background.

Reliability (the P0 train)

  • Every first message reaches its agent, or fleet ls says it did not. A spawn's --prompt, a restart's or a switch's line and a hand-off's were typed with one Enter about two seconds after start, and an Enter that came before Claude Code drew its input box was dropped. A Claude Code agent now gets its first message on its command line, from a file only you can read that its launch deletes; any other binary, and [fleet] first_prompt = "type", has it typed once the box is drawn and checked there. fleet spawn, restart and switch wait up to [fleet] prompt_ack_seconds (45) for the agent's turn to take it and end with ✓ first message taken, or say it was not; --json adds prompt_delivery and prompt_acked.
  • Nothing the fleet types lands in your draft. Every typed write (the manager's nudge, fleet tell, a typed first message, the Remote's paste) reads the input...
Read more

0.8.2 — copies reach the clipboard on macOS, Restart all / Stop all, and no more jumping panes

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 20 Oct 13:48
8e40ad9

A hotfix. A copy in asq reaches the clipboard on macOS. Every agent can be restarted or stopped at once. And a restart no longer takes you off the pane you're on.

0.8.2 shipped as #266. It went through three rounds of independent /code-review (xhigh). Every finding was fixed with a test or answered in its thread. The new UI was also checked twice in a real asq. CI is green on Linux, on both ambient legs, and on Windows.

  • Copy and paste on macOS. In Terminal.app and iTerm2, a copy in asq now reaches your clipboard. 0.8.1 sent every copy only as OSC 52 and said "copied" either way. Terminal.app ignores OSC 52, and iTerm2 drops it unless its clipboard setting is on, so ⌘V pasted whatever the clipboard held before.
    • On a local session, a copy now also goes through the platform's own tool: pbcopy on macOS, clip.exe on Windows and WSL, and wl-copy, xclip or xsel on Linux and the BSDs. It still goes as OSC 52 too, which is the only route over SSH.
    • The toast says "copied" only when the copy can have landed. Otherwise it tells you how to copy instead.
  • Restart all / Stop all. In asq it's ⚙ Settings ▸ Fleet ▸ Restart all… or Stop all…. On the command line it's aisquare fleet restart --all [--managers] and aisquare fleet stop --all [--managers].
    • Both cover every running agent in every project, or only the managers.
    • The dialog lists the agents by project, its confirm button names the count, and Cancel has focus.
    • Exited agents are left alone, and so is an agent that exits while the run goes on. When one agent fails, the rest still run.
  • No more jumping panes. A restart that finishes while you're on another agent's pane leaves you there, and a toast names the restart instead.
  • A click guard. For a second after the view under the pointer changes, Stop and Restart ignore a mouse click aimed at what was there before, and they don't take the focus. A key press is never ignored.
  • Mac-aware help. When you type on a Mac, the help screens name the keys a Mac actually has: c and ctrl+c, ⌘V, and Option-drag or Fn-drag for the terminal's own selection.

Upgrading:

  • From 0.8.0 or 0.8.1, run aisquare upgrade. From 0.7.0 or earlier, re-run the one-line installer.
  • No store migration (the schema stays at v32) and no new dependencies.

The full list, with every detail, is in CHANGELOG.md → 0.8.2.

Known issues: none of these blocks 0.8.2.

  • Suppose an agent's view is on screen and the agent is restarted from outside it, for example by fleet restart --all in a shell. The view stays on the old pane until you click the new row in the sidebar.
  • Under GNU screen reattached over SSH, a copy may still say "copied": screen doesn't say where it was last attached from.

Full diff: v0.8.1...v0.8.2

0.8.1 — the fleet on your phone, and a quieter asq

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 19 Oct 13:49
11704bc

The fleet on your phone: aisquare remote, through your own ngrok tunnel. And a quieter asq: projects you can name, one ⚙ Settings, and an Accounts page that reads at a glance.

0.8.1 shipped as #259. It brings in #243 (Remote, by @Aditya-aisq), #261 (a cleaner fleet UI) and #258 (Rename on the Accounts page). It went through 43 rounds of independent /code-review. Every finding was fixed with a test or answered in its thread. CI is green on Linux, on both ambient legs, and on Windows.

  • aisquare remote: one server on 127.0.0.1:8750 serves a page that ships with aisquare-cli. The page sits behind a secret link and a four-word passphrase, and it is read-only until aisquare remote allow-write on.
  • On the phone you get one feed of what needs you, across every project, with notifications. You also get each agent's live screen and conversation, and the board. With writes on, you can answer a prompt, tell an agent something, or stop, restart or switch it.
  • In asq, R opens the Remote panel. It has Remote on and off, the link, a QR code, the passphrase, the devices, the write switch and an auto-off timer. aisquare remote serve runs Remote without the UI, and the guide is docs/remote.md.
  • Projects have nicknames, which you rename in place on the card. The sidebar lists your projects, then ⚙ Settings (App settings, Accounts, Doctor).
  • The Accounts page is an aligned table showing usage per window. You rename an account in place there, and Log in shows where a login is due.
  • Tasks are numbered on their board (#1, #2, …), and every task command takes the number.
  • A launch copies a folder's trust from the Claude account that gave it. An agent on another account no longer stops at "trust this folder?".
  • fleet reap removes a finished agent's worktree once its work is pushed. The manager stops and reaps agents whose work is done.

Upgrading:

  • From 0.8.0, run aisquare upgrade. It keeps your install's extras. To use Remote, install the new remote extra as docs/remote.md says. From 0.7.0 or earlier, re-run the one-line installer.
  • segno, a small pure-Python QR library, is a new core dependency.
  • config.toml now holds only your changes. On first load, any value equal to 0.8.0's default follows today's default. So an explicit pick = "default" or on_limit = "wait" moves to the new defaults, headroom and switch. To keep the old behaviour, run aisquare config set accounts.pick default and aisquare config set accounts.on_limit wait after upgrading.
  • Nothing becomes a project on its own any more. The directories 0.8.0 captured stay in the store but are no longer listed in asq. Add one you use with aisquare project onboard <path>, or run aisquare config set projects.capture true to keep 0.8.0's behaviour.
  • The store migrates to schema v32 on first open, from 0.8.0's v25, in seven steps. 0.8.0 still opens a v32 store as it is.
  • "name" in project list --json, fleet ls --json and Remote's project list is now the project's nickname when it has one. A script that wants the folder should read root.
  • Remote keeps its state in ~/.aisquare and writes it only once Remote is used. aisquare uninstall --purge removes it with the rest.

The full list, with every detail, is in CHANGELOG.md → 0.8.1.

Known issue: before replacing the install, aisquare upgrade and aisquare uninstall name any live fleet agents. They don't yet check for an aisquare remote serve running in another terminal.

Full diff: v0.8.0...v0.8.1

0.8.0 — a first run you can follow, from an empty home to a manager and two coders

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 13 Oct 15:43
3cebd35

A first run you can follow, from an empty home to a manager and two coders,
and an aisquare that upgrades and uninstalls itself.

0.8.0 is a feature release, shipped as #257. It went through thirteen rounds of independent /code-review (xhigh), rounds 17 to 29. Every finding was fixed with a test or answered in its thread. CI is green on Linux, on both ambient legs, and on Windows.

  • asq opens on Welcome: a project, Claude Code connected, then a manager and
    two coders, all from the keyboard.
  • aisquare upgrade and aisquare uninstall, also as Update and Uninstall in
    asq's Doctor.
  • A Claude Code plugin gives aisquare's memory hooks with no installer:
    /plugin install aisquare@aisquare-cli.
  • aisquare doctor has a row for each coding agent, and with no Node the
    codebase snapshot reads "off (optional)" instead of three warnings.
  • The README is a one-page front page, on GitHub and on PyPI, with a GIF of
    the Welcome walkthrough; its long form moved into docs/.

Upgrading:

  • From 0.7.0 or earlier, re-run the one-line installer, or run
    UV_PYTHON_DOWNLOADS=automatic uv tool install --force --python 3.13 --with tiktoken --refresh-package aisquare-cli 'aisquare-cli>=0.8.0'.
    Not uv tool upgrade: on a pinned install it silently does nothing. Both
    replace the install and drop its extras; with the command, name yours, as in
    'aisquare-cli[serve]>=0.8.0'. Under a uv cooldown (exclude-newer) that
    still hides 0.8.0, the command stops with "No solution found" instead of
    moving you back. From 0.8.0 on, aisquare upgrade does it for you and keeps
    them.
  • The store migrates to schema v25 on first open, from 0.7.0's v24. The step
    adds two columns to fleet_agent: socket_path and boot_id, the socket
    each agent's tmux server listened on and the boot it started in, recorded at
    spawn. upgrade and uninstall ask that socket whether the agent may still
    run. An agent started before 0.8.0 recorded neither, so its socket is asked
    where this shell's tmux looks, and it counts as running until aisquare fleet shutdown ends it. Where its socket is gone it counts too, as --forget is
    then the way, unless it started before this boot: a reboot clears /tmp.
    0.7.0 opens a v25 store as it is, and the agents it starts record neither
    column.
  • aisquare agents connect codex and … cursor now exit 1 with "can't connect
    … yet", instead of reporting a connection that installed nothing. One 0.7.0
    recorded no longer reads as connected; aisquare agents disconnect codex (or
    cursor) clears it.

The full list, with every detail, is in CHANGELOG.md → 0.8.0.

Known issues: none of these blocks 0.8.0, and all are tracked for a follow-up PR in this list on #257. Most are fleet edge cases where a tmux socket file is removed while its server keeps running.

Full diff: v0.7.0...v0.8.0

0.7.0 — accounts, project groups and destinations, and a fleet that survives restarts

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 26 Sep 00:57
53385d4

Accounts, project groups and destinations, and a fleet that survives restarts.

0.7.0 is a feature release. It gathers the whole usability backlog merged through #203: 21 pull requests and 18 issues. Everything in it went through independent review rounds, then a whole-release review whose findings were checked by adversarial verifiers, before it shipped. CI is green on Linux (Python 3.11–3.13), on both ambient legs, and on Windows.

The full list, with every detail, is in CHANGELOG.md → 0.7.0.

Install / upgrade

# fresh install: works out what your machine has, installs what's missing
curl -fsSL https://raw.githubusercontent.com/AISquare-Studio/aisquare-cli/main/install.sh | sh

# already installed with uv (the installer's default)
uv tool upgrade aisquare-cli          # or, if you pinned a version:
uv tool install --force aisquare-cli@0.7.0

aisquare --version    # aisquare 0.7.0
aisquare doctor

Highlights

Claude accounts (#145, #146, #152)

  • Add several Claude accounts and give them a default and a priority order.
  • Spawns go to the account with usage headroom.
  • A usage limit hands the agent over to another account, and the agent keeps its task claims.
  • The usage table shows the reset time and date for both the five-hour and weekly windows.

The fleet UI

  • A resizable navigator (#137). Drag the divider, or use < > =. The width is remembered.
  • Project groups, pins and a manual order (#140), by drag and drop or keys, with undo.
  • Captured is not shown (#139). Directories that hooks only touched stay out of the list until you onboard them. project list --all shows everything.
  • An agent's pane behaves like a terminal:
    • mouse buttons reach programs that track the mouse (#148)
    • select and copy (#120)
    • alt+letter chords (#117)
    • a bare modifier press is silent (#151)
    • a full hotkey audit (#147)
  • Windows are born the width they're shown at (#149), so Claude Code's diff panel fits.
  • The 🔔 rings only for a real prompt (#153).

Agents that survive

  • Restart is the same agent (#138, #144). The launch spec is recorded at spawn and replayed. Stop and Restart are on the agent view, and fleet restart resumes the session.
  • aisquare fleet shutdown (#121) is the fleet's off switch. Rows are no longer stuck at "unknown".
  • Safer hand-overs:
    • they are compare-and-set, so exactly one replacement starts
    • a failed replacement never releases claims it doesn't own
    • a row that outlived its tmux server never acts on another agent that reused its pane id
  • A spawned agent is told the task it was spawned for (#116).

Explainability

  • Pick where a project's traces land with your sign-in (#142): workspaces, studios, use. The key is minted for you, and a minted key is always revoked or kept on record until its revocation is confirmed.
  • A workspace key per project (#141). It is stored at mode 600 and bound to one deployment. A project's destination never falls back to another deployment's gateway, and one project's use never re-points another.
  • Workspace credits (#143) in status, whoami, the Accounts page and doctor --live.
  • The proxy lane is judged by where it actually ships (#131). Setup is a form on the Explainability tab.
  • Auto mode's classifier failure behind the proxy is named (#150): in doctor, at spawn and on refusal.

Store and platform

  • Schema v24. Every migration from v15 up is idempotent. A convergence pass repairs stores another build stamped, such as a v15 or v17 branch build.
  • doctor names what a store lacks instead of calling it "readable".
  • project forget --purge clears every table that references the project.
  • Windows (#65):
    • green CI
    • credentials written atomically under a lock, owner-only from creation, with the DACL read back to check it
    • icacls and whoami run by absolute path
  • Hermetic test suite (#79). The suite no longer reads the real ~/.claude account.

Upgrade notes

  • The first run of 0.7.0 migrates ~/.aisquare/context.db to schema v24. To keep a copy first:
    sqlite3 ~/.aisquare/context.db ".backup ~/.aisquare/context.db.pre-0.7.0"
  • aisquare init --reinit on a config.toml it cannot read now needs --yes. UTF-16 and BOM configs, such as those written by PowerShell 5.1, are read normally.
  • On tmux older than 3.5, shift+enter is sent as C-j. The installer recommends tmux 3.5 or newer.

Known issues

  • Explainability destinations and per-project keys are the newest area. Please report anything odd with aisquare doctor --live --json.
  • A missing project key file makes launches fall back to the machine key (the Explainability tab names the missing file). This is under review.
  • On slow Windows runners, a few wall-clock performance tests can exceed their budgets.

Full diff: v0.6.0...v0.7.0

0.6.0 — the fleet UI

Choose a tag to compare

@AnmolMajithia AnmolMajithia released this 04 Sep 15:01
f4c3387

The fleet UI: bare asq opens one view over every project, agent and
session.
This is the first release where the CLI has a front door — a
two-pane, mouse-driven terminal UI over your projects, the manager agent in
each, and the agents that manager spawns, every one of them a real Claude Code
session you can type into. Everything it does is still a plain command with
--json, and both halves below it — memory and orchestration — work exactly as
they did without ever opening it.

Shipping ahead of feature-complete on purpose, to make internal testing easier;
known gaps are listed in docs/plans/fleet-tui.md and land as 0.6.x.

Added

  • The fleet: asq with no arguments opens one view over every project,
    agent and session.
    A two-pane, mouse-driven UI — a navigator of projects,
    the agents running in each and a Doctor summary on the left; onboarding, a
    project's manager (an agent you task in prose, which plans, spawns and
    steers the others), any agent's real Claude Code session, the board and
    doctor findings with their fixes on the right. Never a chat relayed through
    us: agents run as windows of a per-project session on a private tmux server
    (tmux -L asq, bundled config, tmux ≥ 3.2), so they outlive the UI and
    aisquare fleet attach shows the same session from any terminal. New fleet
    group — spawn · ls · status · tell · stop · attach · reap · rename · pause · resume, --json everywhere — and an explicit ui command. Fleet roles
    manager (the planner with fleet authority), tester (the fleet's name for
    runner) and reviewer (read-only PR review), which launch accepts too.
    Store schema v11: fleet_agent, and a per-project codename
    (adjective-animal, deterministic from the project id; fleet rename
    changes it) that names the tmux session and the fleet/<codename>/…
    branches. A [fleet] config section in which every value is a default —
    permission mode auto for every role, a worktree per coder and reviewer,
    four agents per project, F12 as the escape key, Claude's native agent
    teams off in fleet launches — overridable per spawn or
    in config. Scripts, pipes and --json callers of bare aisquare still get
    usage and exit 2. User guide: docs/fleet.md; the plan and its decisions
    log: docs/plans/fleet-tui.md. Delivered in phases (plan §9); the plan's
    Decisions log records what has landed.
  • CI runs the suite against a machine that looks like a developer's. The
    check job installs .[dev] into a pristine runner — no ~/.aisquare,
    nothing listening on any port, no optional extra — while anyone who followed
    the setup guide has all three. Three fixtures in this repo were green here and
    red for them, and the third was introduced while fixing the second, which is
    what moves this from "be careful" to "have a job".
    • Two variants, because the leaks need opposite proxy states. Measured in
      CI, not assumed: with both regressions reintroduced on a throwaway branch,
      proxy-down errored on the three test_runbook_json_paths.py tests and
      proxy-up on test_json_stdout_is_empty_or_parseable[proxy-down] — disjoint,
      neither catching the other's, while all three check variants stayed green.
      One configuration would have covered half the class and looked like it
      covered the class.
    • Both variants assert their own premise, on both sides of the suite, using
      aisquare explainability status rather than a curl at a literal port: it
      probes whatever the config says, so it cannot drift from what the suite
      reads, and it checks the service/mode contract rather than "something
      answered 200". A variant whose distinguishing condition never held, or which
      lost it midway, is a job reporting the other variant twice — so it fails
      instead, before or after the suite, saying which.
    • The ambient environment has one definition, exported through
      $GITHUB_ENV: AISQUARE_HOME plus the explainability variables
      tests/conftest.py goes out of its way to clear, whose own comment notes an
      operator's shell has them sourced. A per-step home could be dropped from the
      step that runs pytest, degrading the job to check twice, green. Deliberately
      NOT a job-level env: block — that was tried and rejected: runner.temp is
      unavailable there and GitHub rejects the whole workflow file for it, a
      near-silent failure that yields a run with zero jobs and no log, and reads
      from outside as "CI has not started".
    • The variant name is checked against a closed set before anything
      dispatches on it. Every dispatch is [ "$AMBIENT" = "proxy-up" ], so any
      other value — including the empty string, which is what deleting the export
      or renaming the matrix key produces, since an unknown matrix property
      expands to "" with no error — is silently proxy-down, and the matrix reports
      both of its names having run one ambient.
    • It reuses tests/proxy_stub.py rather than inlining a server: probe_proxy
      checks service and mode, so a stub is a contract, and the copy nobody
      runs locally is the one that drifts.
    • Two of the three conditions above are reproduced, and the job says so: the
      extra cannot be installed where the suite runs (it shadows an editable
      checkout), so package covers that axis at import level instead.
  • The packaging job installs the extra. pip install "aisquare-cli[explainability]" is the line the setup guide gives people and
    nothing tested it. The SDK shares this package's top-level import name, so both
    distributions land in one site-packages/aisquare/ and the last writer wins
    the shared __init__.py — a build that read __version__ off that module died
    at import with the extra installed, and no job would have caught it. The job
    asserts both packages import and names the six explainability subcommands,
    because --version cannot distinguish a build that has this integration from
    one that does not.
  • tests/test_ci_covers_the_ambient_environment.py pins all of the above. Its
    checks assert the order and placement of things inside the job rather than
    the presence of a string in the file, because an independent review mutated the
    literal version and found five ways to make the job vacuous with every guard
    green — drop the home from the step that matters, hardcode the stub's port,
    delete either premise assertion, run pytest ahead of the listener, or invert
    the variant condition so the two run under each other's names. Each of those is
    caught now, and the edits a maintainer would legitimately make — reordering the
    variants, adding a third, rewriting the matrix in block style — do not cause a
    false failure.

Changed

  • aisquare serve runs on mcp 2.x. The serve and dev extras require
    mcp>=2.1,<3 (was >=1.10,<2). mcp 2.0.0 renamed FastMCP to MCPServer and
    deleted mcp.server.fastmcp, the module this CLI imported, so the Dependabot
    bump (#73) went red on mypy and the <2 pin was the only thing keeping a
    fresh install green. The port is confined to services/mcp_server.py, the
    serve dependency guard, and the two test files that drive them; the nine
    tools, their wording, and both transports' behaviour are unchanged.
    • The error-wording contract survives, on the seam the SDK now provides.
      mcp 2 still folds a tool's ToolError into Error executing tool <name>: <msg>, so the handler that unwraps our own message back out is still
      needed. It moves from the removed _mcp_server.call_tool() decorator to
      add_request_handler("tools/call", …) on _lowlevel_server, which is what
      the SDK's own migration guide names for replacing a protocol handler (and
      what the SDK itself uses to wrap this method for extensions). A remote
      agent still sees error: reopen requires a note (the feedback), verbatim,
      as an isError result — tests/test_serve.py asserts every one of those
      strings end-to-end through a real client session.
    • A crashed tool is now logged server-side. New in mcp 2.1, not chosen
      here: the SDK tells a crash apart from a deliberate failure by type
      (UnexpectedToolError) and keeps the crash's detail off the wire, so the
      agent sees Error executing tool <name> and nothing else. In 1.x the
      message rode along in the result, which was the only place it went. The
      replacement handler logs the traceback on the server (stderr, which is
      never the protocol channel on either transport) where the SDK's own handler
      would have, so a bug in a tool is still readable somewhere. A rejected
      argument set is the caller's mistake, not a crash, and is not logged as
      one. test_a_crashed_tool_is_an_error_result_logged_server_side pins both
      halves: nothing of the exception on the wire, all of it in the log.
    • HTTP transport settings moved off the server object: host is passed to
      streamable_http_app(), whose only use for it is deciding whether loopback
      DNS-rebinding protection auto-enables (it does, as before), and the port is
      uvicorn's alone, as it already was.
    • The serve guard probes mcp.server.mcpserver, and its message for an
      incompatible major points the other way now — a 1.x is the one that cannot
      work — with pip install 'mcp>=2.1,<3'. The distribution-versus-module
      distinction it was written for (#55) is exactly what makes a 1.x a
      sentence rather than a traceback. It tells majors apart, not minors: the
      pin is what keeps a 2.0.x out, and pip reports that at install time.
    • tests/test_serve.py drives the server through mcp.client.Client(server),
      the in-memory replacement for the removed
      create_connected_server_and_client_session, and reads is_error: field
      names are snake_case in 2.x.
    • Also inherited from 2.x: synchronous tool bodies run on a worker thread
      rather than inline on the ev...
Read more