Skip to content

Deploy your first AI Daemon

AI Daemons are always-on AI runtimes. Unlike an AI Job — which runs on demand or on a cron and then stops — a daemon is a persistent VM that stays up, keeps its state, and serves an interactive control dashboard you can open in your browser. They are billed per hour of uptime (in HollowHost tokens) rather than per run.

This guide walks you through creating, deploying, and operating your first daemon.

Daemon vs. AI Job

  • AI Job — stateless, event-driven function. Triggered manually or by cron, runs to completion, then idles at zero cost.
  • Daemon — stateful, always-on runtime (an EC2 instance). Reachable via a control dashboard, billed continuously while running.

1. Open the Daemons section

From the dashboard sidebar, click Daemons. You land on the daemons list (empty on a fresh account). Click New Daemon to start the wizard.

2. Choose a runtime

The first wizard step picks the runtime image your daemon will run:

  • Hermes (recommended) — full web dashboard with 30+ integrated tools.
  • Pi-agent — interactive terminal via ttyd, remote shell access, real-time logs.

3. Pick a default model

Select the LLM the daemon uses by default for its conversations. Available models in this environment:

  • Claude Sonnet 4.5 (default) — balanced quality/speed.
  • Claude Haiku 4.5 — faster and more economical.

You can change the model later from the daemon's settings.

4. Choose a plan

The plan fixes the VM size and the token burn rate (tokens consumed per hour of uptime). All plans run on arm64 (Graviton) instances.

Plan vCPU / RAM Storage Tokens / hour EBS backups
Hobby 0.5 vCPU · 512 MB 8 GB SSD 150 —
Standard (recommended) 1 vCPU · 1 GB 8 GB SSD 300 —
Pro 2 vCPU · 2 GB 10 GB SSD 600 ✅ weekly, 90-day retention

Pro tier — automated EBS backups

Only Pro daemons get automated disk backups: a weekly snapshot (Sundays 02:00 UTC) of the daemon's volume, retained for 90 days. Hobby and Standard daemons are not backed up — see the pricing FAQ.

5. Name and create

Give the daemon a human-friendly name, then click Create. HollowHost now provisions the runtime:

  1. Allocates a dedicated, isolated execution role for this daemon.
  2. Launches the EC2 instance in the daemon network and pulls the runtime image.
  3. Boots the control dashboard on the daemon's port.

Status flow: DEPLOYING → RUNNING (or ERROR — open the daemon page to read why).

6. Reveal the control password

On first deploy, HollowHost generates a control password and shows it to you exactly once in a modal. Copy it now — it is never stored in plaintext and cannot be retrieved later.

One-time secret

The control password is shown once at deploy time and never again. If you lose it, you'll need to redeploy the daemon to generate a new one. The password protects the public control dashboard, so treat it like any other credential.

7. Open the dashboard

Once the daemon is RUNNING and the dashboard has answered its readiness probe, the Open dashboard button activates on the daemon detail page. It opens the daemon's control URL; authenticate with the control password from the previous step.

8. Operate the daemon

From the daemon detail page you can:

  • Stop / Start — stop preserves the disk (and its state) and halts token burn; start resumes the same VM.
  • Restart — reboots the runtime in place (same instance, same IP).
  • Delete — tears the daemon down and releases its AWS resources.

Token consumption, the consumption trend, and the projected time-until-tokens-out are shown live on the page.

Next steps

Got stuck? File an issue on the GitHub repo or reach out to the HollowHost team.