Squelch

Coding agents read a lot of command output, and most of it is noise. The Squelch pane makes the proxy on your Mac cut that noise out of the command results the machine's agents send their model — passing tests, download and build progress, git's transfer counters, repeated log lines — while keeping errors, failures, warnings and summaries.

The setting belongs to one machine (a workspace, in the settings editor). It is in the editor's Models group. The feature itself — what each family of commands loses, what the model sees, how the savings are counted — is explained in Squelch — Less Noise, Fewer Tokens.

Squelch

Trim the noise out of command output before the model reads it turns the feature on for this machine. It is off by default. It applies from the agents' next request — no restart — and to every agent on the machine.

The same switch is in each session's header: click the Squelch icon after the token count.

What each family of commands loses is in the table in the chapter.

Saved so far

Once the machine's agents have run commands, the pane shows what Squelch saved: the tokens saved, the share of command output left out, the number of outputs sent and trimmed, the commands that saved the most, and a chart of the last two weeks. Tokens are estimated as characters ÷ 4.

The pane ends with a link to rtk — Rust Token Killer (rtk-ai), Apache-2.0, which Squelch is modelled on.