Overview
How much tool output Kompactor shortened before Claude read it, and what that saved you.
What do these numbers mean?
- Tokens saved
- Tool output Claude would have read in full (command results, files it opened), minus what Kompactor handed it instead. Failures, errors and the code Claude needs are always kept.
- Worth
- What those tokens would have cost on the Claude API at the price of the model each session ran on: once when first sent, then again on every later request in the session that would have re-read them, until the conversation was compacted. On a Pro or Max plan it shows the value, not money off your bill.
- Output shortened
- How much smaller the tool output was, all together, after Kompactor trimmed it.
- Replies
- Each time Claude finished answering you. A reply was helped when Kompactor shortened at least one output in it.
Claude's replies
Before and after
Biggest wins
Where savings come from
Left as is
Delete history
Replay your history
Configuration
Choose how hard Kompactor compacts. One level sets everything: toward detail, Claude sees more of the raw output; toward savings, less. Fine-tune any setting below.
What Kompactor does
Advanced settings
Every setting the level controls, explained. Changing one overrides the level for it; "Reset" goes back to the level's value.
In Claude Code
How savings are shown. These don't depend on the level.
explore|general. Empty means every subagent. Applies while "Apply rules to subagents" is on.All configuration
Every setting at every level, what you're using now, and how settings combine.
Code graph
Functions, classes, imports and calls, indexed on this machine so Claude finds code without reading whole files.
Projects
Give a repo its own level. Each repo's languages are detected automatically from its files (pyproject.toml, go.mod, package.json, Cargo.toml, pom.xml and others).
Projects
Setup
The Claude Code CLI and the VS Code extension read the same settings file, so one switch covers both.