shad

A terminal agent that stays interruptible.

Describe a task. shad opens a live shell, makes a plan, runs commands, answers prompts, and accepts corrections while it works.

How shad worksInteraction sequence
  1. 01
    Describe the task

    Give shad an outcome in plain language.

  2. 02
    Review the plan

    See the route before commands run.

  3. 03
    Steer while it works

    Correct the task without discarding the session.

  4. 04
    Stop when needed

    Return control to the person at the keyboard.

Documented behavior. Private alpha.

Control stays at the keyboard.

01

Visible plan

The intended route is readable before the work disappears into a command stream.

02

Live shell

The agent works in the terminal session already in front of the user.

03

Mid-flight steering

A correction can change the task without throwing away useful progress.

04

Immediate stop

The person at the keyboard can end the agent’s run and return to the shell.

Can an agent move quickly without taking control away from the person at the keyboard?

shad is currently a private alpha. There is no public install or signup while the interaction model is still being tested.