Celistra vs Supervisor (supervisord): Process Supervision Comparison

Supervisor (supervisord) has been doing one thing well since 2004: keep a Linux daemon running. Define a process in /etc/supervisor/conf.d/myservice.conf, run supervisorctl start myservice, walk away. It's the answer to "I need this Python script to stay up forever."

Celistra is a different shape — same family of problem, different design choices.

Feature matrix

Where Supervisor wins

Simplicity for traditional service supervision. If you have a Python script that should run forever, an INI file and supervisorctl are 5 lines of config and zero JavaScript. Hard to beat for single-host service supervision.

Battle-tested. 20 years of production use. Edge cases are debugged.

No daemon-per-machine cost. Supervisor is one daemon per host; Celistra is the same architecturally. Tied.

Where Celistra wins

Fleet view across machines. Five Linux boxes with Supervisor = five separate Supervisor instances + your own dashboard glue. Celistra ships the dashboard.

Sandbox + capabilities. Supervisor runs whatever you tell it to with the user's permissions. Celistra wraps spawns in Seatbelt (macOS) or bwrap (Linux) with a capability profile derived from a token bundle. Workspace-only writes; per-process exceptions.

Audit log. Supervisor's logs are stdout/stderr per process. Celistra has a hash-chained audit log of every spawn / kill / capability grant / sandbox block — verifiable, queryable, exportable.

Mobile. Supervisor has no mobile story. Celistra has native iOS + Android with haptic permission approvals.

AI-agent-shaped. Restart-on-crash with awareness of model 500s, capability tokens designed for "this agent can take screenshots but not write outside workspace," browser PTY with multi-client fanout for late-joining clients.

The "supervisor alternative for ai agents" question

Supervisor wasn't designed for AI agents. It works fine for "keep this process up" but the surrounding concerns — what did the agent do, can I approve actions from my phone, is the agent contained — aren't its job. Celistra was built with those as primary requirements.

Migration: Supervisor → Celistra

# Supervisor /etc/supervisor/conf.d/scraper.conf
[program:scraper]
command=python3 /home/me/scraper/run.py
autorestart=true
user=me
stdout_logfile=/var/log/scraper.log

# Celistra (via dashboard or CLI)
celistrad ctl spawn --name scraper --auto-restart \
  --workspace /home/me/scraper \
  'python3 run.py'

Same supervision; Celistra adds workspace boundary, audit log, fleet view, mobile approvals.

Use both

Reasonable. Supervisor for the system services that should stay up forever (databases, web servers, queue workers). Celistra for the AI-agent / multi-machine / interactive supervision tier. Two daemons on a single host don't conflict.