any AMD card.
serves LLMs.

Flash the image. Boot the box. Chat. After that it does not need the internet. Network is only for extra models and attaching nodes to a fleet.

RX 580RX 68007900 XTXMI210APUs

download github

cameod

:9090

vulkan

amdgpu

ready

01how it works

Download. Plug it in. Set an account. Play.

Cameo is an appliance, not a toolchain. The USB already carries the OS. Install copies that image onto disk — it downloads nothing. After reboot the box serves on your LAN with the cables pulled.

Download

One ISO, while you still have a network. Universal if you're unsure; lite if you know the card is Vulkan-only.

Plug in

Flash a USB, boot the AMD machine. Install Cameo to disk is the default menu entry.

Account

Pick a disk, set an admin username and password. A console key is written for you.

Play

Open :9090 from the LAN. A starter model (qwen2.5-0.5b) is already on disk. Chatting does not need the internet.

first boot · tty1

================  Cameo  ================  RX 580 8G · Tier 3 · vulkan OK Web console:  http://192.168.1.40:9090  Console key:  K7m…  Log in as the admin you created.  No internet required to serve.

02what your card does

Three tiers. Vulkan on all of them.

Cameo never silently fails on unsupported hardware. It classifies the card and says so in plain language. Training needs ROCm (tiers 1–2). Inference does not.

1 train + serve7900 XTX · MI210 · official ROCm ROCm, Vulkan fallback

2 serveRX 6800 · 6700 XT · HSA override inference; training community-tested

3 vulkanRX 580 · APUs · no ROCm path inference only

first boot · tty1

================  Cameo  ================GPU 0  Radeon RX 580 8G  pci  0000:01:00.0  vram 8192 MiB  arch gfx803  tier ● Tier 3   no training  why  gfx803 has no usable ROCm path:        Vulkan-only inference, no training.-----------------------------------------Web console:  http://192.168.1.40:9090

That's the real first-boot printout for a tier-3 drawer card — the tier in plain language, the reason, then the console URL. Check yours with cameo gpu-status; the tier is a smart default, not a verdict.

03the console

Administer it from a browser.

cameod starts on boot and binds :9090 with a key. GPUs, endpoints, a playground — on your LAN. Drop a GGUF, chat. No cloud, no extra web stack.

cameod · forge-01 · 0.0.0.0:9090

GPUs & tiers

RX 7900 XTX T1

VRAM24 GBbackendrocm

RX 6800 T2

VRAM16 GBbackendvulkan

RX 580 T3

VRAM8 GBbackendvulkan

Endpoints

qwen2.5-0.5brunning · 127.0.0.1:8080 · vulkan · /v1 · offline

04one card → a cluster

Same two commands on 1, 4, or 9 cards.

One box is image and flash — no Wi-Fi, no extra models required. Connecting boxes to each other is the only time you join a network: Ethernet if you have it, otherwise iwctl on that node so it can attach to the fleet.

05get it

Image. Flash. That’s the whole path.

Get an .iso, write it to a USB, boot the AMD box. No toolchain, no Wi-Fi, no extra models. A starter is already on the image. Network is later — extra GGUFs, or attaching this box as a node in a fleet.

1 · Download the image

Take the file named cameo-*.iso (or .iso.part* if it was split). Ignore “Source code” zip/tar — those are not flashable. If the newest tag has no ISO yet, use the last tag that lists one.

Universalany AMD card

Vulkan always; ROCm only when the card can train. Unsure? This one. Pick a release that includes an .iso.

releases — pick an .iso

Litesmaller · vulkan only

Known old card (RX 580, APU). Last flashable lite image:

v0.1.0-beta.2 ISO

2 · Flash a USB

Windows: Rufus in DD Image mode (ISO mode often will not boot), or Etcher. Linux/macOS: Etcher, or:

sudo dd if=cameo-*.iso of=/dev/sdX bs=4M status=progress oflag=sync

3 · Plug it in, set an account

Disable Secure Boot (or the stick is ignored and Windows boots). Put the USB in the AMD machine. Install Cameo to disk is already the default entry. It shows the detected GPU and tier, then asks for:

which disk to erase admin username admin password

Nothing is downloaded. The installer copies the live image onto that disk, writes a console key, and enables cameod. Type the disk name to confirm, then reboot and pull the USB.

4 · Play, offline

From a laptop or phone on the same LAN (plug the box into Ethernet if it has a port), open the URL printed after login — http://cameo.local:9090 or the IP. Enter the key. Start qwen2.5-0.5b (smoke test, already on disk). Chatting never phones home. Extra GGUFs: drop them in /var/lib/cameo/models.

Fleet: that’s when you connect Wi-Fi

A single box does not need Wi-Fi. Join a fleet of nodes — then give that node a network so it can attach:

Ethernet — plug in, DHCP is automatic Wi-Fi — iwctl on that node only extra models — cameo pull

iwctl
  station list
  station wlan0 scan
  station wlan0 get-networks
  station wlan0 connect 'YourSSID'

Then cameo fleet / hub URL on that node. One machine, one card, one house: stay air-gapped after you flash.

Developers: container or build the ISO yourself

Container — host owns the GPU driver:

podman build -f containers/Containerfile -t cameo:vulkan .
podman run --rm -p 9090:9090 \
  -v cameo-models:/var/lib/cameo/models \
  --device=/dev/kfd --device=/dev/dri \
  --group-add video --group-add render \
  cameo:vulkan

Build the ISO on an Arch host:

git clone https://github.com/sadlowskik/Cameo
sudo ./scripts/build-iso.sh
sudo CAMEO_EDITION=lite ./scripts/build-iso.sh

Beta, pre-v1. Flash an .iso, not Source code. Check SHA256SUMS. The disk installer erases the drive you pick. Source: github.com/sadlowskik/Cameo.

any AMD card.serves LLMs.