aish - the terminal your agent can afford to read
by•
AI coding agents run terminals built for humans. Every git diff, test run, or build log comes back full of banners, passing-test spam, repeated paths, and stack trace noise .
aish sits in front of your shell commands and gives agents compact, evidence-preserving observations instead. Same commands, same information that matters just without the noise.
What command wastes the most context for your coding agent? Tell us below — that's exactly what we're building toward next.
Replies
Hey PH 👋
I built AgentShell because I kept watching my coding agent burn context on stuff no one actually needs to read — full test suite output for one failing test, entire diffs when only two files actually changed, build logs three thousand lines long for one root-cause error.
aish wraps the commands you already run (aish npm test, aish diff, aish npm install) and gives the agent back only what matters: the failing test, the exact hunk, the actual error — nothing else.
It's a CLI, install is one line: npm install -g aish, then aish init in any repo to get your agent using it automatically.
Would genuinely love to know: what's the single command that wastes the most tokens for your agent right now? That's directly shaping what I build next.
If you could add a way to mark certain patterns as always-strip (like my project's specific ANSI color codes), it'd save me from tweaking the filter rules every time. Right now I'm rebuilding the same ignore list for different repos.
@aysunvcyf That's a great point, and a common enough pattern (ANSI codes, project-specific noise) that it's worth a real fix rather than per-repo tweaking. I was thinking a .aishignore-style file, or a always-strip pattern list in your repo-local config that persists across runs, so you set it once per project instead of rebuilding the filter every time.