AI Task Manager 0.1.6

Install

macOS, Intel and Apple Silicon (one universal build): download AI.Task.Manager_0.1.6_universal.dmg, open it and drag AI Task Manager to Applications. The app is not notarized yet, so macOS blocks the first launch: open it once, then go to System Settings, Privacy & Security, scroll to Security and click Open Anyway (on macOS 14 or earlier, right-click the app and choose Open). Or run xattr -dr com.apple.quarantine "/Applications/AI Task Manager.app".

Windows 10 and 11: download AI.Task.Manager_0.1.6_x64-setup.exe (or AI.Task.Manager_0.1.6_x64_en-US.msi). The installer is not signed yet, so SmartScreen will warn: More info, then Run anyway.

Installs with update checks on are offered this release at their next launch or popover open, or within four hours. Update checks stay off by default; with them off, download and install over the old copy.

Both installers were built by this repository's own GitHub Actions workflow from the tagged commit. Nothing leaves your computer: docs/privacy.md lists every network call the app can make and the commands that prove it.

What is in 0.1.6

Budgets for agents and time spent at the limit, both in all nine languages,
and a fix for client budgets on the 31st.

Added

• A budget for each agent, and an alert for a runaway one. The Agents
  view has a new Budgets section: a monthly dollar budget per agent, picked
  from the agents the app knows, Claude Code's built-in ones included, and
  never typed. The month is the calendar month. Each row shows what the agent
  has spent so far this month, and you get one notification per agent per
  month the first time it passes its budget. Two dropdowns in Settings >
  Notifications watch the sessions running now: a spend per hour (the cost of
  a session's last ten minutes, carried over an hour) and how long a session
  can stay open and still working. Either one gets you one notification per
  running agent per day. Nothing is set by default: no budget row, no runaway
  check and no notification exists until you set a figure. The one thing
  shown before that is a hint under the spend setting, what your own fastest
  session running now is spending, when there is one, so there is something
  to go by. On a flat-rate plan the figures are API-equivalent value, not a
  charge, as everywhere else here. The app only tells you: it never stops,
  pauses or edits an agent. A budget that has been passed and a session past
  your figure are also listed under Worth a look in the Agents view, and as
  unscored rows in the Audit, which open the Agents view. The figures are
  kept in agent_watch.json on your machine and never reach the seat report
  or the local API (docs/privacy.md says exactly what).
• Time spent at the limit. A provider's page says how often each limit
  reached 100 percent in the last 30 days and for how long in all, one line
  per limit, most time first, for example "Weekly: fully used 3 times in the
  last 30 days, 5h 7m in all". A limit that never got there has no line. It is
  worked out from the readings the app already keeps. Time counts only
  between two readings taken within 90 minutes of each other, up to the reset
  when one falls between them, so a stretch when the app was not running is
  left out rather than guessed and the figure is a floor. When one limit adds
  up to two hours or more, an unscored row in the Audit says so and opens the
  Usage tab.

Changed

• The TLS library is updated to rustls 0.23.45, which clears
  RUSTSEC-2026-0285. A TLS 1.3 peer could send handshake messages at the
  wrong encryption level without being refused. The handshake was still
  authenticated, so this is closer to a correction than to an exposure, but
  rustls is what the update check and the agent's push to a collector speak
  through.
• Dependencies are audited every week. A new workflow runs cargo audit
  and npm audit on the dependencies that ship, whenever a lockfile changes
  and every Monday, because an advisory is published against code that did
  not change. It fails on a known vulnerability and only warns on a crate
  that is unmaintained, unsound or yanked.

Fixed

• A client's month to date was short on the 31st. It was cut out of a
  30-day series, so on the 31st of a month with 31 days the 1st fell one day
  outside it: the figure was short by whatever the 1st had spent, and a
  budget the month had really crossed could go unreported on the month's last
  day. That is seven months a year, one day each. The scan now adds up the
  calendar month itself, so the figure is whole on every day.
• A server's 30-day call figure could be another server's. Claude Code
  logs a call to a server whose name ends in punctuation, such as foo!,
  under the shorter name foo, so a configured server called foo was shown
  the other's calls and result sizes as its own. A server whose calls cannot
  be told apart from another's now shows no figure, and so does the one it
  would have been counted under, rather than a number that may be someone
  else's.
• The version in a seat report and in the HTTP user-agent was stuck at
  0.1.0. Both took it from a crate whose number was never bumped, so every
  release since 0.1.1 said 0.1.0. They now say the app's own version, and a
  test fails a release that bumps one number and not the other. A seat on an
  older build keeps reporting 0.1.0 until it updates.