Deploy Options

Pick where GAIA runs. All options need about 8 GB of RAM, which is what drives the price.

Pick one

Prices are list prices from 2026-10-14. Check the provider before you buy.

Option	Setup	Status	Monthly cost
Linux server	Run `gaia up` over SSH	Tested	About $44 to $48 (DigitalOcean, Lightsail), less on Hetzner, $0 on Oracle Always Free
DigitalOcean	Click Create Droplet	Not listed yet	$48
Coolify	Paste one Compose file	Not tested yet	Your server
Render	Deploy a Blueprint	Not tested yet	About $252, plus Cloudinary
Railway	Create 9 services by hand	Not tested yet	Usage-based, plus Cloudinary

A Linux server is cheapest and is the path we test. Render and Railway bill per service, so the same 8 GB costs several times more.

Linux server

Create the server

Use 4 vCPU and 8 GB RAM. Allow ports 80 and 443 (or 3000 without a domain).

Follow the walkthrough

Run the steps in the walkthrough over SSH.

Add a domain (optional)

Point the domain’s DNS A record at the server, then:

gaia up --domain gaia.example.com

DigitalOcean Marketplace

Once listed, create a Droplet from the GAIA listing (8 GB / 4 vCPU or bigger), wait for first boot, then open http://<Droplet IP>/setup. The image comes with Docker and the CLI, makes its own secrets, and only allows SSH, HTTP and HTTPS.

Maintainers: building and submitting the image

The Packer template lives in deploy/digitalocean/, in the format of DigitalOcean’s marketplace-partners repository. It runs their 03-force-ssh-logout.sh, 90-cleanup.sh and 99-img-check.sh at a pinned commit.

Install Packer and the plugin: packer plugins install github.com/digitalocean/digitalocean.
Create a DigitalOcean API token and export it: export DIGITALOCEAN_TOKEN=....
From deploy/digitalocean/, run packer build marketplace-image.json. The build ends with 99-img-check.sh; fix every [FAIL] before submitting.
Create one Droplet from the snapshot and check that http://<IP>/setup loads and sudo -iu gaia gaia doctor passes.
Submit the snapshot in the Vendor Portal with the listing copy, the quick-start text from files/etc/update-motd.d/99-one-click, and links to these docs. If you have no Vendor Portal login, apply at marketplace.digitalocean.com/vendors.

Coolify

Uses deploy/coolify/gaia.yaml.

Create the resource

In your Coolify project choose New → Docker Compose Empty, paste the contents of the file, and save. Use a server with at least 4 vCPU and 8 GB of RAM.

Set the domain

Under Domains, set the domain for the proxy service. Coolify issues the certificate and generates the passwords.

Add your AI key and deploy

Optionally set OPENAI_API_KEY in Environment Variables, or add a provider at /setup later. Deploy, then open the domain and go to /setup.

Not yet tested on a live Coolify. If a service fails, its logs and GET /health/ready show which one. Please report what you find.

Render

Uses the Blueprint render.yaml. It needs a Cloudinary account, because a Render disk cannot be shared between services (BLOB_STORE=cloudinary).

Create the Blueprint

Click the button below, or in the Render Dashboard choose New → Blueprint and connect the GAIA repository (or your fork), then apply it.

Render asks for the values marked sync: false. Enter your CLOUDINARY_CLOUD_NAME, CLOUDINARY_API_KEY and CLOUDINARY_API_SECRET, your OPENAI_API_KEY (or leave it empty), and pending for the other three for now.

Fill in the private addresses

Render gives every private service a hostname with a random suffix, and a Blueprint cannot build a URL from it. Once the services exist, copy each internal address from its Connect panel and set these on gaia-backend (the worker reads them from there) and gaia-web:

Service	Variable	Value
`gaia-backend`	`MONGO_DB`	`mongodb://<gaia-mongo address>/gaia`
`gaia-backend`	`RABBITMQ_URL`	`amqp://guest:guest@<gaia-rabbitmq address>/`
`gaia-web`	`API_BASE_URL_INTERNAL`	`http://<gaia-backend address>/api/v1/`

Each address looks like gaia-mongo-a1b2:27017. Save, and Render redeploys.

Open GAIA

Open the gaia service’s onrender.com URL (or add your domain to it) and go to /setup.

Not yet deployed end to end. If a service fails, check its logs and GET /health/ready on the gaia URL. Please report what you find.

Render makes the secrets once and keeps them. Services use :latest images. To pin one, put a tag like v0.7.0 in each image URL.

Railway

There is no template yet, so you create the services yourself. You also need a Cloudinary account, because a Railway volume belongs to one service (BLOB_STORE=cloudinary).

Service list and variables

Create these in one project (New, Docker Image, or GitHub Repo for the proxy). Then add a public domain to gaia on port 8080.

Service	Source	Volume	Variables
`gaia`	This repo, `RAILWAY_DOCKERFILE_PATH=deploy/proxy.Dockerfile`	—	`GAIA_SITE_ADDRESS=:8080`, `GAIA_BACKEND_UPSTREAM=${{gaia-backend.RAILWAY_PRIVATE_DOMAIN}}:8000`, `GAIA_WEB_UPSTREAM=${{gaia-web.RAILWAY_PRIVATE_DOMAIN}}:3000`
`gaia-web`	`ghcr.io/theexperiencecompany/gaia-web:latest`	—	`API_BASE_URL_INTERNAL=http://${{gaia-backend.RAILWAY_PRIVATE_DOMAIN}}:8000/api/v1/`
`gaia-backend`	`ghcr.io/theexperiencecompany/gaia:latest`	—	the shared variables below, plus `WORKER_TYPE=main_app`
`gaia-worker`	`ghcr.io/theexperiencecompany/gaia:latest`, start command `python -m arq app.worker.WorkerSettings --custom-log-dict app.workers.config.log_config.ARQ_LOG_CONFIG`	—	the shared variables below, plus `WORKER_TYPE=arq_worker`
`mongo`	`mongo:8`	`/data/db`	`GLIBC_TUNABLES=glibc.pthread.rseq=0` (needed on Linux 6.19+)
`postgres`	`postgres:18-alpine`	`/var/lib/postgresql`	`POSTGRES_USER=postgres`, `POSTGRES_PASSWORD=<generate>`, `POSTGRES_DB=langgraph`
`redis`	`redis:8-alpine`, start command `redis-server --appendonly yes`	`/data`	—
`rabbitmq`	`rabbitmq:4.1.8-management`	`/var/lib/rabbitmq`	`RABBITMQ_NODENAME=rabbit@localhost`
`chromadb`	`chromadb/chroma:1.5.1`	`/chroma/chroma`	`PERSIST_DIRECTORY=/chroma/chroma`, `CHROMA_SERVER_HOST=0.0.0.0`, `CHROMA_SERVER_PORT=8000`

Shared variables for gaia-backend and gaia-worker (Project Settings → Shared Variables, then reference them):

ENV=selfhost
AUTH_MODE=local
PYTHONPATH=/app/apps/api
MONGO_DB=mongodb://${{mongo.RAILWAY_PRIVATE_DOMAIN}}:27017/gaia
REDIS_URL=redis://${{redis.RAILWAY_PRIVATE_DOMAIN}}:6379
POSTGRES_URL=postgresql://postgres:${{postgres.POSTGRES_PASSWORD}}@${{postgres.RAILWAY_PRIVATE_DOMAIN}}:5432/langgraph
CHROMADB_HOST=${{chromadb.RAILWAY_PRIVATE_DOMAIN}}
CHROMADB_PORT=8000
RABBITMQ_URL=amqp://guest:guest@${{rabbitmq.RAILWAY_PRIVATE_DOMAIN}}:5672/
# A Railway volume attaches to one service, so files go to Cloudinary.
BLOB_STORE=cloudinary
CLOUDINARY_CLOUD_NAME=
CLOUDINARY_API_KEY=
CLOUDINARY_API_SECRET=
# openssl rand -hex 32 for each; never change them after first boot.
AGENT_SECRET=
BOT_SESSION_TOKEN_SECRET=
EMAIL_UNSUBSCRIBE_SECRET=
SHARE_GRANT_SECRET=
GAIA_BOT_API_KEY=
SANDBOX_EXECUTE_TOKEN_SECRET=
INSTANCE_SECRET=
OPENAI_API_KEY=

Not yet deployed end to end on Railway. Check GET /health/ready on your domain after the first deploy.

After you deploy

Open /setup

Go to your address followed by /setup: http://<server IP>:3000/setup for gaia up without a domain, http://<Droplet IP>/setup, or your Render, Railway or own domain.

Create the admin account

The first account is the admin. Use a password of at least 8 characters.

Pick an AI provider

Paste a key for OpenAI, OpenRouter or Gemini, or use a local Ollama. Skip it if you set the key during deploy.

Web search, bots and integrations are optional. Add them later in Settings, Providers & Tools; see the walkthrough.

To	Linux server or Droplet	Render, Railway
Check health	`gaia doctor`	`GET /health/ready` on your URL
Update	`gaia update`	Change the image tag, redeploy
Back up	`gaia backup`	Disk snapshots plus `mongodump` and `pg_dump`; see Backups

Get Started

Configure

Connect Apps

Use It Everywhere

Maintain

Reference

Pick one

Linux server

DigitalOcean Marketplace

Coolify

Render

Railway

After you deploy

​Pick one

​Linux server

​DigitalOcean Marketplace

​Coolify

​Render

​Railway

​After you deploy

Pick one

Linux server

DigitalOcean Marketplace

Coolify

Render

Railway

After you deploy