local AI operations

Your agents keep moving after you leave.

Advanced, local-first automation for long AI runs. Observe every session, route the safe decision, and preserve the record.

relay builder / doff-01
$ canvas.empty / select an agent

automation registry

Small tools. Serious leverage.

Numbered as they ship. Every entry is used on real work before it is offered here.

selected / 01

doff

operational

A local supervisor for Claude sessions that stops blocking prompts and quiet windows from becoming dead runs.

sweep
~0.4s
runtime
local
record
JSONL
platform
macOS
$doff runwatching

01private testing

session supervisor / macOS

doff keeps the run alive.

It watches Claude sessions across the terminal and desktop app, distinguishes quiet from stuck, takes bounded action, and records exactly what happened.

doff / sessions

supervisor active

03 visible 01 action taken sweep 0.4s
?
defaultblocked on confirmation
create tests/run_suite.py? 4m answered / yes
↻
validatorrunning test suite
python3 test-doff.py live observing
·
builderidle, no composer text
last activity 21m ago 21m nudged
12:04:18decision appended → ~/.switchboard/doff-log.jsonlok
01

Observe

Enumerates the live surfaces on your machine and refuses a leg that did not answer cleanly.

02

Distinguish

Separates blocked, active, idle, signed-off, and unknown states before it does anything.

03

Act

Answers known confirmations, invokes commands correctly, or nudges a genuinely stalled run.

04

Prove

Leaves a durable local record instead of treating a send receipt as evidence that work moved.

testedClaude Code 2.1.233
testedDesktop 1.30096.5
runtimemacOS / local

Distribution becomes a one-line install when private testing closes. Versions are stated because accessibility surfaces move.


next action

Bring the job you keep doing by hand.

Request doff, propose the next automation, or report a problem. The message goes to a person and the record stays durable.