Skip to main content

FAQ

Why not cloud AI and a few SaaS apps?​

  • Each app sees only its own data. A chat assistant doesn't know your notes unless you paste them in.
  • Each one is a separate account and subscription, and can change under you.
  • Here everything works on one folder, workspace/: code-server with claude and opencode, automated by decree, using one model URL.
  • With Local AI, OpenViking indexes workspace/ and Honcho remembers you across sessions.
  • It runs on your hardware. No subscription.

Is it hard to set up?​

Hostnames, TLS, secrets, logging and the automation engine are set up once for every service. Setup is ./existential.sh, then ./existential.sh flow on for each thing you want. Most automations are configured in a cron file's frontmatter, not code: see the four knobs.

Is it only for homelab people?​

The flows are chores:

What if I don't have a GPU?​

Core doesn't need one. For AI flows, answer external when asked and point EXIST_MODEL_URL at a machine that has one, or use the CPU tier. See the hardware questions.

Do I have to turn everything on?​

No. Every service is a flag, and a disabled one writes nothing. The first run builds Core (work dir, terminal, automation, notifications). See Enable/Disable Services.

What happens to my data if I stop using it?​

It's plain files on your disks, readable without Existential running.

Is it finished?​

No. Services are added, replaced, or moved to the graveyard. The flows documented here all run today.

Can the stack's own agent change the stack?​

Not recommended. Use a coding agent against a frontier model, on the machine with the repo.

  • opencode can't drive Hermes (why).
  • A local model at 64k context is too small for this repo.
  • Hermes works on workspace/: the agent automations that write into workspace/ai/.