Quickstart
This guide takes you from nothing to a working AION in about 15 minutes, on a Linux server or on a laptop with Docker. It is meant for IT people who want to try the product; no knowledge of the code is needed. For a production installation with your own domain and HTTPS, follow Deploy with Docker afterwards.
What you need
- A machine with Docker and Docker Compose (Linux, macOS, or Windows with WSL2). For a trial a small server or a recent laptop is enough, because the AI model normally runs elsewhere; size the machine up only if you also host the model on it.
bashandcurl.- Access to the AI model you want to use (its address and, if required, its API key).
1. Install
Run the installer. It downloads the ready-made images (no source code needed), creates a folder aion-agent/ with the configuration and starts everything:
curl -fsSL https://raw.githubusercontent.com/AION-by-ASA-Computer/AION_Agent/main/scripts/install.sh | bash
Useful options (append after bash -s --, for example ... | bash -s -- --domain aion.example.com --email it@example.com):
| Option | Meaning |
|---|---|
--domain <name> | Public domain name. AION then gets an HTTPS certificate automatically (Let's Encrypt). Without it, AION answers on plain HTTP at http://localhost. |
--email <address> | Contact email for the HTTPS certificate. |
--dir <path> | Installation folder (default ./aion-agent). |
--version <tag> | A specific AION version (default: the latest). |
When it finishes, these addresses are available:
| Address | What it is |
|---|---|
http://localhost/ | Chat app |
http://localhost/admin | Admin panel |
http://localhost/api/health | Health check of the API server (answers {"status": "ok"}) |
http://localhost/docs/ | This documentation |
With --domain, replace http://localhost with https://<your domain>.
2. First login and AI model
- Open the admin panel (
/admin) and sign in with useradminand passwordadmin. - You are asked to change the password. Do it now: this is the only administrator account.
- The First setup wizard opens. The important step is the first one, LLM: enter the details of your AI model (provider, address, key), click Test connection, pick the model from the list. The next steps (embeddings, OCR, web search, policy, SSO) can be left at their defaults and changed later from Settings.
- At the end of the wizard AION restarts the services it needs to apply the configuration.
3. Send the first message
- Open the chat app (
/) and sign in with the sameadminaccount. - Write a message, for example "What can you do?", and press Enter. The answer appears as it is generated.
- Try a file: drag a PDF into the chat and ask "Summarize this document in 5 bullet points".
To add colleagues, go to Admin panel → Users (see the Administrator checklist).
4. Make the first API call (optional)
- In the admin panel → Users, edit your user and turn on Can create API keys.
- In the chat app → Settings → API keys, create a key with the Chat permission and copy it (it is shown only once).
- Call the agent:
curl -s http://localhost/api/v1/chat \
-H "Authorization: Bearer aion_XXXXXXXX_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" \
-H "Content-Type: application/json" \
-d '{"message": "Hello! Answer in one sentence."}'
The answer is a JSON object whose text field contains the reply. The API guide continues from here.
Stop, restart, update
Run these commands from the installation folder (aion-agent/):
docker compose -f docker-compose.ghcr.yml ps # status of the services
docker compose -f docker-compose.ghcr.yml logs -f backend # API server logs
docker compose -f docker-compose.ghcr.yml down # stop (data is kept)
docker compose -f docker-compose.ghcr.yml up -d # start again
To update to a new version, use the Update now button in the admin panel (when available) or run the installer in upgrade mode; both make a backup first. See Self-update.
Something went wrong?
| Symptom | What to check |
|---|---|
| The page does not open | docker compose -f docker-compose.ghcr.yml ps: all services should be running/healthy. The first start can take a couple of minutes. |
| The chat answers with an error about the model | The LLM configuration: Admin panel → Settings → LLM, run Test connection again. |
| HTTPS certificate not issued | The domain's DNS A record must point to this server, and ports 80 and 443 must be reachable from the internet. |
More answers in the FAQ.