Self-hosting

Every Decosa app runs from the same containers on your own NVIDIA GPU. Nothing is sent to our servers and there are no Decosa charges. The local service speaks the same API as the hosted one, so apps only need a new base URL.

Publishing soon. The container images (${DECOSA_REGISTRY}/decosa-*) and the compose file are not public yet. Commands below use placeholders in angle brackets until they are.

Hardware

Linux x86_64, a recent NVIDIA driver, and tens of GB of free disk for model weights.

Install

1. Check the GPU, Docker, and the NVIDIA Container Toolkit:

   nvidia-smi
   docker compose version
   docker run --rm --gpus all ubuntu nvidia-smi
   

   If the last command fails, install the NVIDIA Container Toolkit, then run sudo nvidia-ctk runtime configure --runtime=docker and restart Docker.

2. Fetch the compose file and start:

   mkdir -p ~/decosa && cd ~/decosa
   curl -fsSL "${DECOSA_COMPOSE_URL}" -o compose.yaml
   docker compose pull
   docker compose up -d
   

3. Wait for health, then point your app at the local base URL:

   curl -fsS http://localhost:<PORT>/healthz
   

Each use-case page has a prompt that has a coding agent do all of this, including installing Docker and the toolkit if they are missing.

Patient data stays on site

• Do not expose a scribe box to the internet: no port forwarding and no public tunnels. Reach it from the clinic network or a private VPN.
• Put TLS in front of it if other machines use it. Keep disk encryption on.
• Receipts on a self-hosted box are signed by the box's own key (status attested). The key is created on first start in the data volume; back it up, and publish its public key (/attest/signing-key) wherever people will check your records. It is an attestation by your box, not a proof that the model ran.