Choose a Deployment

DocsGPT runs as an API (which also serves the web UI), a Celery worker, PostgreSQL and Redis. Every install method below runs those same pieces. They differ in who manages them, where the data lives, and which version you get.

At a glance

Method	Best for	What runs	Address	Data	Version
Installer / `docsgpt up`	Most people: a laptop, a home server or a cloud VM	Docker Compose stack managed by the `docsgpt` command	`http://localhost:7091`, or `https://<domain>` with `--domain`	Named Docker volumes; settings in `~/.docsgpt/server/.env`	The installed package’s release
Standalone Compose file	Running the images with plain Docker Compose, no Python on the host	The same stack, from one downloaded file	`http://localhost:7091`	Named Docker volumes; `.env` next to the file	`latest` unless `DOCSGPT_IMAGE_TAG` is set
Source checkout, `setup.sh`	Trying changes, building your own images	Compose files under `deployment/`	`http://localhost:5173` (UI) and `:7091` (API)	Uploads and indexes in `application/` in the checkout, the database in the `docsgpt-oss_postgres_data` volume; `.env` in the repository root	`develop` (follows `main`) unless `DOCSGPT_IMAGE_TAG` is set
pip install	Hosts without Docker, or your own process manager	`docsgpt api` and `docsgpt worker` (or `docsgpt up --native` services) against your Postgres and Redis	`http://127.0.0.1:7091`	`~/.docsgpt/server`	The installed package’s release
Kubernetes	Clusters, several API or worker replicas	Manifests in `deployment/k8s/`	Through `kubectl port-forward` or your Ingress	Postgres volume, S3 for uploads	Pinned in the manifests
Air-gapped	Networks without internet access	Any of the above, with images and models brought in	As above	As above	Whatever you mirror
Development environment	Working on DocsGPT itself	Postgres and Redis in Docker, the app from source	`http://localhost:5173`	The checkout	Your working tree

Which one to pick

Start with the installer. curl -fsSL https://docs.ac/install | bash (or irm https://docs.ac/install.ps1 | iex on Windows) installs the docsgpt package and runs docsgpt up. It generates the secrets, keeps the stack on the release you installed, and gives you docsgpt status, docsgpt backup and docsgpt upgrade. See the Quickstart.
On a cloud VM, such as Amazon Lightsail, Hetzner, DigitalOcean or any other provider, use the same installer with a domain. Point the domain’s DNS record at the VM, open ports 80 and 443, then run:
```
curl -fsSL https://docs.ac/install | bash -s -- --domain docs.example.com
```
Caddy obtains the certificate and DocsGPT asks for an access token. Keep the other ports closed. For separate user accounts, add OIDC single sign-on.
Prefer plain Compose when you manage containers yourself, for example with an existing Compose-based setup. The standalone file runs the same images as docsgpt up, and docsgpt up --adopt can take it over later.
Use a checkout to change the code or build images. setup.sh and the checkout Compose files are meant for local use: they run develop images by default and publish on 127.0.0.1. Set DOCSGPT_IMAGE_TAG in .env to pin a release.
Use pip where Docker is not available or not wanted. You provide PostgreSQL and Redis; docsgpt up --native can run the API and worker as launchd or systemd services on macOS and Linux.
Use Kubernetes when you already run a cluster and want replicas, rolling upgrades and your own Ingress.

⚠️

Before DocsGPT is reachable by anyone else, set AUTH_TYPE, a real INTERNAL_KEY and ENCRYPTION_SECRET_KEY, and publish it over HTTPS. docsgpt up does most of this for you. Work through the security checklist for any other method.

After installing

App Configuration: where .env lives and the main settings.
docsgpt CLI Reference: every docsgpt command.
Upgrading: how to move to a new release on each install method.