Skip to main content

Self-host Nora on your own infrastructure

Run Nora on your own infrastructure with the one-line installer or Docker Compose. Covers system requirements, install paths, and opening the dashboard.
Nora is self-hosted by design. You run the full stack — Nginx, Next.js frontends, Express API, PostgreSQL, and Redis — on infrastructure you control. This page covers system requirements, the two install paths, and how to reach the dashboard once the stack is running.

System requirements

Before you install, make sure your host meets the following requirements.
The setup script checks for Docker, Docker Compose 2.24.4 or newer, Git, and OpenSSL and installs missing prerequisites before proceeding. Standalone docker-compose v1 is not supported because Nora’s hardened overlays use modern Compose merge tags.
The installer script is the fastest path to a running Nora instance. It handles dependency checks, secret generation, access mode configuration, optional admin account creation, local port selection, and stack startup in one interactive pass.
1

Run the installer

Open a terminal and run the command for your platform.
2

Follow the interactive prompts

The installer walks you through each configuration decision in order:
  1. Verify or install Docker, Docker Compose, Git, and OpenSSL
  2. Clone the repository (if running via the curl pipe)
  3. Generate or preserve secrets for JWTs, AES encryption, managed backups, Agent Hub keys, and the database
  4. Choose platform mode: self-hosted (operator-defined resource limits) or PaaS (Stripe billing)
  5. Choose whether to keep the GA Docker backend enabled and whether to opt into the experimental Proxmox LXC target. Kubernetes clusters are registered after setup in Admin -> Kubernetes. Proxmox stays disabled by default and requires secure API/SSH configuration plus the real-host smoke test.
  6. Choose access mode: local only (http://localhost:8080), public domain behind a proxy, or public domain with TLS at Nginx
  7. Optionally create a bootstrap admin account
  8. Start the stack; after login, add your LLM provider key in Settings
When the installer finishes, it writes .env and starts the stack with docker compose up -d. On macOS, Linux, and WSL the file is restricted to mode 0600. On Windows, setup removes inherited access and grants access only to the current user, SYSTEM, and local Administrators; setup stops if it cannot secure the file.
In local mode, setup checks host ports before startup. If 8080 is busy, it offers a free web port such as 8081 and updates NEXTAUTH_URL. If 127.0.0.1:4100 is busy, it offers a free backend API port and writes BACKEND_API_PORT.
3

Open the dashboard

The installer prints the base URL it configured. For local mode, the default is:
If setup selected a different port, use the printed URL instead. If you skipped the bootstrap admin account, go to /signup on that origin. On a fresh self-hosted instance the signup page runs in “claim this server” mode and the first account you create becomes the platform admin (see Security architecture). Hosted PaaS cannot skip bootstrap-admin creation.

Reconfigure vs clean reinstall

If you run setup again from an existing checkout, it asks which maintenance mode you want:
Reconfigure mode intentionally preserves the existing DB_PASSWORD from .env because the Postgres Docker volume is already initialized with that password. If you manually replace .env with a different DB_PASSWORD while keeping volumes, the backend cannot sign up or log in users until the password is restored or the volume is recreated.

Manual setup

Use manual setup when you want full control over every configuration value before the stack starts.
1

Clone the repository

2

Run the setup script from inside the repo

Running bash setup.sh from inside a cloned repo skips the clone step and takes you through the same interactive configuration flow as the one-line installer.
If you prefer to configure everything by hand, copy the example environment file instead and edit it directly:
3

Start the stack

Once .env is configured, run the non-interactive update path once. It materializes the gitignored Compose secret files, refreshes the hardened overlay, and starts all services:
On Windows, use pwsh -File .\setup.ps1 -Update. After this first materialization, ordinary docker compose up -d commands reuse the protected files.Advanced Linux/macOS operators who intentionally bypass setup can run bash scripts/materialize-compose-secrets.sh .env before docker compose up -d. Re-run the helper whenever any of the six core values changes.The production overlay materializes the core JWT, encryption, Agent Hub, backup, and database credentials as read-only files under /run/secrets for backend and worker containers. Marketing receives an explicit allowlist of only its runtime/OAuth values, not the complete deploy env.
Do not start the stack with placeholder values for JWT_SECRET or ENCRYPTION_KEY. Generate each with openssl rand -hex 32 before running the update command. Also replace both DEFAULT_ADMIN_EMAIL and DEFAULT_ADMIN_PASSWORD placeholders with a secure bootstrap account. A self-hosted install may blank both values to use first-account claim through /signup; hosted PaaS requires both values before startup.
4

Open the dashboard

Once the stack is up, open your browser to http://localhost:8080. On self-hosted Nora, open /signup to create your operator account if you did not pre-seed a bootstrap admin. On a fresh database this runs as first-run claim, and the first account created becomes the platform admin. Hosted PaaS requires a pre-seeded administrator.

Kubernetes (Helm)

Run Nora itself on a Kubernetes cluster with the public OCI chart at oci://ghcr.io/solomon2773/nora. Its source lives in infra/helm/nora. The chart deploys the full control plane (nginx edge, the three frontends, backend API, provisioner and backup workers) with in-chart PostgreSQL and Redis by default, or external data stores via values.
This resolves the latest published chart. For a reproducible production install, run helm show chart oci://ghcr.io/solomon2773/nora, record its version, and pass that value with --version. The chart refuses to install without real secrets — there are no insecure defaults. New installs also require distinct workspace/API and Agent Hub hash secrets; older upgrades retain their shared hash-key fallback until you provide the dedicated Agent Hub value. The selected Secret is mounted at /run/secrets only in control-plane and PostgreSQL pods; frontend pods do not inherit it. For a public deployment set publicUrl, enable ingress.*, and terminate TLS at your Ingress controller.
A cluster-hosted control plane cannot use the local Docker target (ENABLED_BACKENDS=k8s) because its pods have no Docker socket. It can still use operator-registered Remote Docker targets over SSH when Nora is self-hosted and the control-plane pods can reach the private remote-host network. Register agent clusters under Admin → Kubernetes and mount their kubeconfigs with kubeconfigs.existingSecret. In-place release upgrades from the Admin UI are a Docker Compose feature — use the Helm upgrade command from the chart README instead.
See the chart README for all values, design notes, and the local Kind smoke test.

Accessing the dashboard

After installation, the following URLs are available in local mode. Public-domain mode uses the same paths on your configured origin.
If you configured a public domain (for example https://app.example.com), replace http://localhost:8080 with your domain in all of the paths above.
Nora dashboard immediately after the stack comes up

What happens next

Once you can reach the dashboard, the next steps are:
  1. Go to Settings and add an LLM provider key (Anthropic, OpenAI, Google, or another supported provider).
  2. Go to Deploy, name your agent, choose a runtime mode, set resource limits, and click Confirm & Deploy Agent.
  3. Open the agent detail page to verify it is running, test Chat, inspect Logs, and open Terminal.
Settings — LLM provider keys section with masked entries See the quick start for a full walkthrough of the first 15 minutes with Nora. To keep the Nora control plane on this installation while running agents on another private Docker server or VM, follow Connect a Remote Docker host. The walkthrough covers SSH registration, host-key pinning, firewall boundaries, deployment validation, and safe retirement.

Useful Docker Compose commands