Installation with Docker Compose

The setup assistant writes a complete, working configuration with every secret generated. You start it, put HTTPS in front, open the ports, and then check each layer from the outside in.

Step 1: make the files

Run the setup assistant in an empty folder:

mkdir mnema-talk && cd mnema-talkdocker run --rm -it --user "$(id -u):$(id -g)" \
  -v "$PWD:/out" ghcr.io/mnema-xyz/mnema-talk setup

-it lets it ask you questions, --user makes the files yours instead of root's, and -v is the folder it writes to. No clone and no build are needed.

It asks for the address members will open (for example https://chat.example.com), the Owner's user name (default admin), the language of the default channel names, and optionally this machine's address in your network, mail settings and the one-click update button. Then it checks what it can see from where it runs: whether the name points to this network's public address, the clock, and whether the line is behind CGNAT. It ends with a numbered list of what to do next.

It does not overwrite an existing .env: new secrets would not fit an existing database.

Without a terminal session

For scripts or a NAS web interface, pass everything as flags. setup --help lists them all; among them are --local for a trial, --lan-ip, the --smtp-* settings, --auto-update, and --check, which only lists the settings that are missing or still a placeholder.

docker run --rm --user "$(id -u):$(id -g)" -v "$PWD:/out" \
  ghcr.io/mnema-xyz/mnema-talk setup --non-interactive \
  --public-url https://chat.example.com --admin-username admin --language en

Step 2: start it

docker compose up -d --waitcurl -fsS http://127.0.0.1:8080/api/health

The first start pulls the images, which takes a few minutes. When the health check answers with "status":"ok", the app is running. Nobody can reach it from outside yet: the app port listens on the machine itself only, so the only way in is your reverse proxy.

The stack is three containers: mnema-talk (the app), PostgreSQL and SeaweedFS. Optional parts start only when you enable their Compose profile: autoupdate for the Update now button and turn for an external coturn relay.

Step 3: HTTPS in front

Point the DNS record at your server and set up the reverse proxy: Domain, DNS and TLS. Once https://chat.example.com shows the sign-in page, the web part is done.

Step 4: open the ports

Voice needs two more rules besides HTTPS: UDP 50000–50050 and 3479 on UDP and TCP. Ports and firewall has the table; at home, Behind a router has the forwarding steps.

Step 5: check it layer by layer

Test in this order and stop at the first thing that is not right; one broken layer makes everything after it look broken.

  1. The site. On a phone with Wi-Fi off, open your address. You should see a padlock and the sign-in page.
  2. Live updates. Sign in as the Owner. The page should stay usable and never say “Connection lost”.
  3. Admin console > Status. Wait about twenty seconds after a start. Every amber or red item names the setting to change.
  4. Test my connection. The button in Admin > Status runs in your browser, needs no microphone and tells you whether direct UDP works, only the relay works, or nothing does. Run it from outside your home network. In a terminal, docker compose exec app mnema-talk doctor prints the same checks from the server's side.
  5. A real call. Create an invite, open it on the phone on mobile data, join a Talk from your computer and speak. You should hear each other within about two seconds.

If a step fails, Troubleshooting goes through causes and fixes in the same order.

Step 6: the first sign-in

Sign in with the Owner's user name and the password the assistant printed. Change the password, then delete the ADMIN_INITIAL_PASSWORD line from .env: it only creates the account on the very first start. What to set up next: First steps as Owner.

Try it on one computer

Want to look around before you buy a domain? Answer yes to “Just try it on this computer?” in step 1, or run setup --local. Start it, open http://localhost:8080 and sign in. Two browser tabs with two accounts can join a Talk together.

Plain http:// works only on localhost: on any other http:// address, browsers block the microphone and camera, so you can read and write but not talk. Docker Desktop on Windows or macOS is fine for this trial, but it is not a tested hosting target.

Without the assistant

The compose file and .env.example of a release are the whole deployment. The manual steps, every required variable and how to pin a version: Deployment in the repository. CI runs exactly those steps against every image it builds.

Something wrong on this page? Tell us.