Getting started
Get one assistant replying first. Email, calendar, voice devices, and other integrations can be added when you need them.
What you need
Run ./install.sh on the OE server and keep the one-time first-run credential it prints. Installation requirements and Docker instructions are in the installation guide. If setup is incomplete and you lose the credential, run oe bootstrap on the server. In Docker, use docker exec openensemble node scripts/first-run-bootstrap.mjs.
1. Open OE and create the owner
Open https://<your-server-ip>:3739 in your browser. Replace the placeholder with the address of the machine running OE. The installer creates a self-signed certificate; check that you reached your own server before accepting its warning. A browser on that same machine can also use http://localhost:3737.
Enter the first-run credential, then choose the owner’s name, emoji, and password. To move an existing installation, choose Already have an OpenEnsemble backup? instead. See Backup and restore.
The owner manages this installation. Admins can also manage privileged settings such as providers and public access; regular users use the access granted to them. See Users & access.
2. Connect one chat provider
Open Settings → Providers as owner or admin. Choose one path:
- Cloud API: enter your provider’s API key, enable the provider, and save.
- Local model: enable Ollama or LM Studio and enter the address of the machine running it. Install/load a chat model there first.
- Account login: use an available provider login; each user connects their own account. See LLM providers.
For a local server on another machine, use that machine’s address. localhost means the machine or container running OE. For Docker’s host model server, see the installation guide’s Docker section.
Cortex handles internal memory and reasoning tasks. You still need a chat provider for your assistant. See Which model does what.
3. Create your assistant
Choose Create Your Assistant on the empty chat screen, or open Agents → + New Agent. Give it a name and emoji, choose a model from the provider you enabled, and save. Personality is optional.
New accounts start in Single assistant mode: that first assistant becomes your primary assistant and handles your enabled skills. Existing accounts keep their ensemble setup. To add specialists later, open Settings → Agents → Agent setup → Agent ensemble. See Single assistant and ensembles.
4. Check the first reply
Send “Hello. What can you help me with?” A successful setup produces a reply from your selected assistant. Then try one capability you have enabled. Web search, email, and calendar each need their own setup; a working chat model does not connect those services automatically.
If there are no models to choose or the reply fails, use Troubleshooting.
Add the next thing you need
| Goal | Guide |
|---|---|
| Keep separate conversations | Saved chats |
| Organize an ongoing project | Project spaces |
| Connect mail or Google Calendar | Email & calendar |
| Control home devices | Home Assistant |
| Connect external tools or another AI app | MCP connections |
| Schedule work or notifications | Tasks & scheduler |
| Build a tablet display | Display dashboards |
| Add hands-free interaction | Voice devices |
| Protect and maintain your installation | Backup, restore, auto-update |