Runs on the AI you already pay for

The AI you already use, working even when you aren't.

Tell Karsa what you want — once. It becomes a system on your machine: morning briefs, standing reactions, a backlog worked overnight.

$curl -fsSL https://karsa.app/install.sh | sh

Requires Claude Code or Codex — you probably already have one. No account, no new AI bill. macOS & Linux.

New here? Read the manual or browse the packages.

One sentence, then it runs itself

you · 4:42 PM

brief me on this repo every morning at 7

karsa · 4:42 PM

Done — every day at 07:00 I'll post a brief in this chat. Pause or delete it anytime.

karsa · next day, 7:00 AM

Morning brief: 3 commits since yesterday — the payment-retry fix landed and tests are green. One PR is waiting on your review. CI is clean; nothing overdue.

Illustrative transcript — the chat UI streams real runs live. Behind it: a planner turned your sentence into a standing job; a timer woke your own machine at 07:00; your Claude/Codex subscription did the reading.

Why Karsa

  1. 01

    Persistent — not another chat window

    Most AI tools answer a prompt and forget. Karsa keeps state: standing schedules, reactions to your data, and a task backlog a worker agent executes — with retries, ordering, and progress reported back to your chat. It runs whether or not you're at the keyboard.

  2. 02

    Runs on the subscription you already have

    Karsa is engine-agnostic: it drives Claude Code today and Codex tomorrow, and your systems don't care which. That $20–200/month you already pay sits idle most of the day — Karsa is how it works the other 22 hours. No new AI bill, no per-token markup.

  3. 03

    Yours — on your machine, in plain files

    Everything runs locally. Your systems, schedules, and data are readable files under your home, versioned in git — inspectable, editable, and yours to back up. No account required to start.

  4. 04

    A real runtime underneath

    Karsa is a local-first runtime where AI-powered systems live — think an application server whose primitives are agents, events, schedules, tasks, and storage instead of HTTP and SQL. Engines are interchangeable muscle; the runtime — and everything it holds — is yours. Designed so the same project can lift to the cloud when you scale.