Try VEP free

Fixed

Repeating tasks stop filling your twin's memory

Your twin learns from what a scheduled task finds, not from the same instructions over and over.

  • AI employees
  • Reliability

What you get

  • When your twin ran a scheduled task, for example checking your inbox every 30 minutes, every run stored the task's instructions in its memory again, as if you had just given them. One task added hundreds of such copies in a week. A repeated run no longer does that.
  • A run that ends with nothing to report now adds nothing to your twin's memory. What a run finds, such as a new request in your inbox, and anything you write to your twin while the task is running are still remembered.
  • When you change a task's instructions, the new wording is learned on its next run.

More updates

  1. Fixed

    Reports on team tasks are checked against the tasks themselves

    When your AI employees work as a team, they hand tasks to each other. Until now, the check of what an AI employee says could not see those tasks, so a correct report that it took or closed a task could get a note saying it could not be confirmed. Now the check reads the tasks themselves.

  2. Fixed

    Learning from your chats now waits instead of being dropped

    Your AI employee learns from what you ask it and what it answers in your chat with it. When that cannot happen right away, because of a usage limit, a sign-in problem, an outage or a busy moment, the question and answer now wait and are tried again instead of being dropped, so your AI employee learns them once the problem is gone.

  3. Fixed

    Something learned late no longer replaces your correction

    Your AI employee sometimes learns from your chat later than it happened, for example after waiting out a usage limit or an outage. Now what you said last in the chat stays current: an older statement learned late is kept in its history and no longer replaces a newer correction.

  4. Improved

    Your AI employee's claims are checked against what it did

    When your AI employee tells you in its chat that it did something or that it will do something later, that statement is checked against what it actually did. A check that is busy now runs again a few minutes later instead of leaving the reply unchecked.