MatrixRTC/LiveKit Setup
This guide assumes that you are using docker compose for deployment.
You can find help setting up MatrixRTC in our support room - #continuwuity:continuwuity.org
Instructions
1. Set up your domain
LiveKit should live on its own domain or subdomain. In this guide we use livekit.example.com - this should be replaced with a domain you control.
Make sure the DNS record for the (sub)domain you plan to use is pointed to your server.
2. Set up the LiveKit services
Using LiveKit with Matrix requires two services - LiveKit itself, and a brokering service (lk-jwt-service) that grants Matrix users permission to connect to it.
You must generate a key and secret to allow the Matrix service to authenticate with LiveKit. LK_MATRIX_KEY should be around 20 random characters, and LK_MATRIX_SECRET should be around 64. Remember to replace these with the actual values!
LiveKit provides a utility to generate secure random keys
Create a docker-compose.yml file as following:
Next, we need to configure LiveKit. In the same directory, create livekit.yaml with the following content - remembering to replace LK_MATRIX_KEY and LK_MATRIX_SECRET with the values you generated:
Consult config-sample.yaml for all LiveKit options.
Firewall hints
You will need to allow ports 7881/tcp and 50100:50200/udp through your firewall. If you use UFW, the commands are: ufw allow 7881/tcp and ufw allow 50100:50200/udp.
3. Telling clients where to find LiveKit
To tell clients where to find LiveKit, you need to add your lk-jwt-service's address to the foci field of the [global.matrix_rtc] section of your Continuwuity config file.
The variable should be a list of servers serving as MatrixRTC endpoints. Replace the URL with the address you are deploying your instance of lk-jwt-service to:
If you configure Continuwuity via environment variables, use the following:
This will expose LiveKit information on the following endpoints for clients to discover:
/_matrix/client/unstable/org.matrix.msc4143/rtc/transports(MSC4143 unstable, behind auth)/.well-known/matrix/client(fallback, not behind auth. Only enabled if[global.well_known].clientis set)
4. Configure your Reverse Proxy
Reverse proxies can be configured in many different ways - so we can't provide a step by step for this.
All paths should be forwarded to LiveKit by default, with the exception of the following path prefixes, which should be forwarded to the JWT/Authentication service:
/sfu/get/healthz/get_token
6. Start Everything
Start up the services using your usual method - for example docker compose up -d.
Testing
To test that LiveKit is successfully integrated with Continuwuity, you will need to replicate its Token Exchange Flow. Follow the steps below while checking Docker logs (docker compose logs --follow), in order to help troubleshooting any issues.
First, you will need an access token for your current login session. These can be found in your client's settings or obtained via this website.
Then, using that token, fetch the discovery endpoints for MatrixRTC services:
In the output, you should see the LiveKit URL matching the one configured above.
With the same token, request another OpenID token for use with the lk-jwt-service:
You will see a response as below:
Next, create a payload.json file with the following content:
payload.json
Replace matrix_server_name and claimed_user_id with your information, and <openid_access_token> with the one you got from the previous step. Other values can be left as-is.
You can then send this payload to the lk-jwt-service:
The lk-jwt-service will, after checking against Continuwuity, answer with a jwt token to create a LiveKit media room:
Use this token to test at the LiveKit Connection Tester. If everything works there, then you have set up LiveKit successfully!
Troubleshooting
To debug any issues, you can place a call or redo the Testing instructions, and check the container logs for any specific errors. Use docker compose logs --follow to follow these logs in real-time.
Before debugging, ensure the following prerequisites:
- The user has permissions to send call events into the room. See the Appendix section on call modes for more details.
- WebRTC is enabled, if using a browser client.
Common errors in Element Call UI
-
MISSING_MATRIX_RTC_FOCUS/MISSING_MATRIX_RTC_TRANSPORT: the client cannot find a suitable SFU. This is likely due to LiveKit being missing from Continuwuity's config file, and therefore not being exposed properly. -
OPEN_ID_ERROR: Your client can reach out tolk-jwt-service, butlk-jwt-servicehas problems authenticating you against Continuwuity. In this case, checklk-jwt-servicelogs for more details. The section below on Docker loopback networking issues may be of help. -
Calls fail to start with
ServerUnreachableerror: a LiveKit URL has been properly configured and exposed in Continuwuity, but the client cannot connect to it for some reason.
For browser-based clients, you can inspect connections using DevTools' Networking tab, to see which requests are erroring out. You may also want to toggle on Element Call's Developer tab via Settings > Preferences > Developer mode - various call details and advanced settings can be further tweaked there.
Docker loopback networking issues
Some distros do not allow Docker containers to connect to the host's public IP, and this breaks connections from lk-jwt-service towards livekit and continuwuity if they're on the same host. This can be tested by the following command:
To resolve this, follow the steps in Docker loopback networking issues of the Troubleshooting page.
Incorrect IP addresses for LiveKit
By default, LiveKit automatically finds its public IP address(es), which are reflected in the "Establishing WebRTC connection" section of the connection test page. If these IPs are incorrect, you may want to hardcode your own IP by doing the following:
Run the LiveKit tests again, and the new updated node_ip should be included as one of the candidates.
You may further restrict the advertised address to only the node_ip with the following configurations:
Related Documentation
Guides:
- Element Call self-hosting documentation from element-hq
- Community guide with overview of LiveKit's mechanisms
- Community guide using systemd
Configurations:
- Livekit's
config-sample.yaml- LiveKit configuration file with full options
Specifications:
- MSC4143: MatrixRTC - Real-time communication over Matrix
- MSC4195: MatrixRTC Transport using LiveKit Backend
- MSC3401: Native Group VoIP Signalling - state events for legacy calls
- MSC4075: MatrixRTC invites and notifications - for ringing users
- MSC4354: Sticky Events - for MatrixRTC 2.0 calls
Source code:
Other:
- Matrix VOIP and LiveKit - Community member deep dive on current VOIP solutions on Matrix
Appendix
MatrixRTC calling modes and enabling sticky events
MatrixRTC currently supports 3 calling modes:
- Legacy mode (deprecated): calls are established by sending state events to a room. The call initiator's SFU hosts the entire call for all remote and local participants.
- *Compatibility mode: same as Legacy mode, but utilizes multiple SFUs between federating servers. Clients publish media streams to their own SFUs, and fetch remote streams from remote SFUs. This is the most commonly used mode nowadays.
- MatrixRTC 2.0 mode: also utilizes multiple SFUs, but establishes calls with [sticky events] instead of state events.
To support Legacy and Compatibility mode, ensure that the user is allowed to send the relevant state events. Usually, this means promoting them to a power level of 50 or above. Alternatively, set the power level required to send org.matrix.msc3401.call.member events to 0.
To support MatrixRTC 2.0 mode, sticky events has to be enabled on the Matrix server. In Continuwuity, this is as easy as setting enable_sticky_events = true in the config file.
Creating LiveKit tokens via livekit-cli
To create room tokens directly without going through lk-jwt-service, use the Livekit CLI tool.
First, download the release tarball and extract it. Then:
Additional TURN-over-TLS on 443 configuration
Most of the time, LiveKit does not need TURN to function. However, there are situations where clients may be unable to use nonstandard ports or UDP connections, such as in highly restrictive networks. In such cases, a TURN-over-TLS server on port 443 could relay traffic for the clients.
You can either use LiveKit's built-in TURN server, or integrate LiveKit with coturn.
Using LiveKit's built-in TURN server
First, set up LiveKit's built-in TURN server with its own domain - this example will use livekit-turn.example.com in our example.
Recreate the LiveKit container (with docker compose up -d livekit) to apply these changes. Remember to allow the new 50300:50400/udp ports through your firewall.
Then, configure a route from port 443 on the host back to the livekit-turn.example.com service on port 5349. To multiplex the service and LiveKit's WebSocket on the same port, use a layer 4 reverse proxy with SNI routing capabilities, such as Caddy-L4, on the host system.
Using an external TURN server (coturn)
First, make sure coturn's min-port and max-port do not overlap with LiveKit's port range:
Then, generate a long random TURN secret for LiveKit, and add it to your coturn config under the static-auth-secret option. coturn allows multiple secrets in its configuration, so set a different one for LiveKit to use.
After that, refer to the following TURN-over-TLS on 443 instructions to set up coturn with TLS, as well as multiplexing with LiveKit's WebSocket on port 443.
Then configure LiveKit, making sure to replace COTURN_SECRET with the one you generated:
Restart LiveKit, coturn, and Caddy-L4 to apply these changes.
Once the services are configured, run the Testing steps again to check that TURN-over-TLS on 443 is working. There should now be a green tick in the LiveKit connection test page showing "Can connect to TURN" to confirm this is set up successfully.