---
title: Run Computers on your own daemon
description: "What a daemon needs to host Venom Computers: Docker, RAM and disk sizing from the measured numbers, the venom-computer image (published and digest-pinned, or built locally), the compose profile, a reachable endpoint for the live view, Chromium updates, and Kubernetes."
section: computers
order: 14
audience: [operator]
sources: [docs/adr/0078-venom-computers.md, deploy/computer/Dockerfile, docker-compose.yml, scripts/daemon.sh, scripts/chromium-update.sh, scripts/release/computer-image-ldflags.sh, .github/workflows/release.yml, .github/workflows/chromium-update.yml, deploy/helm/venom/values.yaml, services/go-orchestrator/internal/computers/enable.go]
features: []
---

Computers run on a daemon you own. This page is the checklist for a fresh machine. The one-click switch is described in
[Enable Computers on a daemon](/docs/computers/enable.md); the settings and admission rules are in
[Venom Computers on a daemon](/docs/deploy/computers.md).

## Requirements {#requirements}

| | Needed | Why |
|---|---|---|
| OS | Linux (macOS is not supported yet) | the daemon reaches a host container at its address on the `venom-computers` Docker network |
| Where the daemon runs | on the host, as `venom-daemon` or the orchestrator binary | it starts the browser hosts through Docker, so it cannot be inside a container |
| Docker | installed, running, usable by the daemon's user | the browsers run in `venom-computer` host containers |
| RAM | at least 4 GiB; plan on 194 MiB per active browser (p50, 224 MiB p95) | [measured](/docs/computers/capacity.md#sizing); the daemon budgets 450 MiB until its hosts report real numbers |
| Disk | at least 10 GiB free where the daemon keeps its data | the image and the browser profiles; a fresh profile is 3.7 MiB, and it grows as the agent browses |
| Network | outbound HTTPS to the cloud, and for the live view an endpoint the browser can reach | see [below](#endpoint) |

A 16 GiB budget holds about 84 active browsers, and the daemon can hold 100 seats (hibernated seats cost disk, not memory).
That is an extrapolation of a 20-active run, so check your own pages with `make computers-load` ([Capacity](/docs/computers/capacity.md#run)).

## The image {#image}

A release publishes `ghcr.io/elishahbiz/venom-computer` for linux/amd64 and linux/arm64. The release's `venom-orchestrator`
and `venom-daemon` binaries have the digest of that exact build linked in, and the same reference is in the release as
`venom-computer-<version>.image` (covered by `SHA256SUMS`). **Enable Computers** pulls that digest; nothing else is needed.

To use another image, or one you built, set `VENOM_COMPUTERS_IMAGE` in the daemon's environment. It wins over the digest in
the binary and over the saved setting:

```bash
make computer-image                                  # builds venom-computer:dev from deploy/computer/Dockerfile
export VENOM_COMPUTERS_IMAGE=venom-computer:dev      # the default when a build from source has no published digest
export VENOM_COMPUTERS=on
```

A daemon built from source has no published digest, so Enable builds the image from the source tree when there is one.

## Docker Compose {#compose}

The compose stack's orchestrator runs in a container and has no Docker, so it cannot host browsers. The `computers` profile
only prepares the image, and the daemon itself runs on the host:

```bash
docker compose --profile computers build computer-image     # venom-computer:dev
COMPUTERS=on make daemon CLOUD_URL=https://… JOIN=vjt_…     # the same build, then the usual daemon stack
VENOM_COMPUTERS=on venom-daemon                              # on the host: hosts' data is under $VENOM_DATA_DIR/computers/hosts
```

Back up `$VENOM_DATA_DIR/computers/` (the seat table and the browser profiles) with the rest of the daemon's data.

## A reachable endpoint for the live view {#endpoint}

Watching a computer and taking control need the browser to reach the daemon directly. Give the daemon a `public` or `tunnel`
https endpoint (`VENOM_MESH_ENDPOINTS`); without one the live view falls back to thumbnails through the cloud and **Take
control** is unavailable. See [Watch a computer and take control](/docs/computers/watch-and-control.md).

## Chromium updates {#chromium}

The image pins Chromium from Debian's security archive, and Debian removes a superseded build, so a stale pin fails the image
build instead of shipping an old browser. `.github/workflows/chromium-update.yml` checks Debian's package index and opens a PR
that bumps `CHROMIUM_VERSION` in `deploy/computer/Dockerfile`. It is opt-in: it runs by hand from the Actions tab, and you
uncomment its weekly `schedule` to automate it. `scripts/chromium-update.sh` does the same check on your machine. The next
release builds the image with the new browser and publishes its new digest; update the daemon to get it.

## Kubernetes {#kubernetes}

Docker-in-a-pod is not used. On Kubernetes the orchestrator already runs in a container, so Computers stays off there and the
Helm chart does not turn it on. The design for a later release: the daemon creates each host as a pod through the Kubernetes
API, from the same `venom-computer` digest. That is not built yet. Until then, run a daemon on a Linux machine with Docker
and join it to the same cloud.
