Remote work that survives the network.

Run shells, jobs, and agent tasks through disconnects.

View on GitHub Install

Auto-reconnect Resumable exec Local MCP SSH fallback

onyx

$ interactive shell over unreliable networks

Stable interactive sessions.

Reconnect fast. Pick up where you left off.

Reconnect fast Resume instantly

Remote session stays live SSH fallback when needed

Start once. Detach. Check back later.

onyx exec

status starting gpu-box • job_84f31

$ onyx exec gpu-box --detach -- python train.py

[job] job_84f31 running detached

train epoch=01 loss=0.912 samples=4096

train epoch=02 loss=0.734 samples=8192

[net] connection lost...

[job] resumed seq=27

train epoch=03 loss=0.581 samples=12288

train epoch=04 checkpoint=best.pt

Run onyx exec prod -- ./deploy.sh

Run it live.

Detach onyx exec gpu-box --detach -- python train.py

Leave it running.

Logs onyx logs gpu-box job_xxx

Check progress later.

Agents talk local. Live progress, reconnect, and resume.

Agent local client

Onyx MCP stdio server

Remote Host hetzner-dev

onyx_exec

onyx_jobs

onyx_logs

onyx_kill

target=hetzner-dev · stream=true

brew install shervin9/onyx/onyx --formula

Update

brew upgrade shervin9/onyx/onyx --formula

Connect

onyx user@host

Remote server is provisioned automatically.