Is my AI agent about to compact?

ACTVT shows how full each live agent session’s context is, as a bar and a percentage of the model’s window, in the session list and in the inspector, along with the token count (“153,192 / 200,000 tokens”). If you turn on Context filling in Settings → Notifications and choose a threshold, ACTVT notifies you once when a live session goes past it. After the session compacts, it can tell you again. The window size comes from ACTVT’s own model table, and ACTVT tells you that. It shows how full the window is, but it doesn’t predict when the agent will compact.

0:544 chapters

Step by step

  1. 0:00Turn on the warning

    Open Settings and scroll down to Notifications. Turn on Context filling and pick when you want to hear about it. I’m using 70% for this demo.

  2. 0:17A long run

    This is a real run. Claude Haiku 4.5 is reading five big log files, one at a time, and its context bar is at 16% of Haiku’s 200K window. As it keeps reading, the bar jumps to 77%, which is past the 70% I set.

  3. 0:36ACTVT lets you know

    ACTVT sends a notification saying the context is at 76%.

  4. 0:43The details

    When I open the session, it’s at 153,192 of 200,000 tokens, and it says where ACTVT gets the window size from. Now it’s up to you whether to let it compact, or wrap up and start a fresh session.

Try it on your own agents

ACTVT is free to start, and reads the session files Claude Code, Codex CLI and Pi already write. Nothing to configure.