N6JET Reflectors and Gateways logo: URF, XLX, M17, VEX and C4FM, with a background of binary 0s and 1s

Setting Up a Vexillum (VEX) Reflector


These are my lab notes from building VEXJET, my own Vexillum server at vex.n6jet.com, between 2 and 11 November 2026. They follow the order I did things in, and they include every problem I ran into and how it was solved. If you want to know what Vexillum is first, read Vexillum (VEX) Reflector Explained.

At a glance

Read this first

Vexillum is beta software. Flags, forms and behavior are still changing from one release to the next. Everything here was true for v0.2.0b in November 2026. Newer releases may work differently, so check the docs at vexdv.io before you copy anything. Never put real passwords on a web page or in shared notes.

Contents

  1. Plan before you install
  2. Download and install
  3. The config file
  4. Run it as a service
  5. First start and the admin page
  6. Add the modes (instances)
  7. Put the API and dashboard on the web
  8. List it on DVRef
  9. List it on VexDV
  10. Hotspots and host files
  11. Bridges
  12. Joining the modes (converter)
  13. D-STAR and the hardware vocoder
  14. All the lessons in one list
  15. Sources

1. Plan before you install

Most of my problems came from sharing one server with other reflectors. If Vexillum has a server to itself, you can skip much of this section.

What VEXJET ended up using:

ModeVexillum defaultVEXJETWhy it changed
DMR620306203162030 is used by xlxd and urfd
M17170001700217000 is urfd, 17001 is mrefd
YSF4200042000Free on VEXJET's own address
NXDN4140041400Free on VEXJET's own address
P254100041100Took over the address of a retired P25 reflector
D-STAR (DExtra)3000130001Free on VEXJET's own address
Admin, metrics, public API8080, 9090, 80818080, 9090, 80858081 was already used by another program

2. Download and install

Releases are on the author's own code server, linked from vexdv.io Downloads. There is no GitHub mirror. When the code server was down for a day, I simply had to wait. I emailed the author, and it was back the next day.

  1. Download vexillum-linux-amd64 and SHA256SUMS.txt. For v0.2.0b the file is 21,020,834 bytes.
  2. Check the checksum before you run anything: sha256sum -c against the SUMS file.
  3. Make a system user with no login shell, for example vexillum, and a folder /opt/vexillum owned by it.
  4. Copy the program to /opt/vexillum/vexillum, owned by root, mode 755. It has only two flags: -config and -version.

3. The config file

The config file is small. Mine, with comments removed:

[runtime]
name = "VEX-JET"

[logging]
level = "info"
format = "text"

[database]
url = "sqlite://vexillum.db"

[admin]
enabled = true
listen = "127.0.0.1:8080"

[metrics]
enabled = true
listen = "127.0.0.1:9090"

[public_api]
enabled = true
listen = "127.0.0.1:8085"
rate_limit_per_minute = 120
rate_limit_burst = 20

Lessons

4. Run it as a service

I wrote a systemd unit that runs Vexillum as the vexillum user with the usual hardening (NoNewPrivileges, ProtectSystem=strict, ProtectHome). Two extra tricks were worth it on a shared server.

Trick 1: a port lockdown

Vexillum may try to open default ports that other reflectors already hold. systemd can block that. I allow only the ports VEXJET should use and deny everything else:

SocketBindAllow=tcp:8080
SocketBindAllow=tcp:9090
SocketBindAllow=tcp:8085
SocketBindAllow=udp:62031
SocketBindAllow=udp:17002
SocketBindAllow=udp:42000
SocketBindAllow=udp:41400
SocketBindAllow=udp:41100
SocketBindAllow=udp:30001
SocketBindDeny=any

Put one port on each line. On systemd 255 a space-separated list on one line was silently ignored. Each time you add a mode, add its port here, open it in the firewall for that IP address only, and restart.

Trick 2: replies from the right address

The M17 form has no address field, so Vexillum listens for M17 on every IP address of the server. Packets came in on VEXJET's address, but the replies went out from the server's main address, and hotspots behind a home router threw them away. Rewriting the reply address in the firewall did not work: the replies vanished. What did work was routing by user. Everything the vexillum user sends uses a routing table whose source address is VEXJET's IP:

ExecStartPre=+/usr/sbin/ip route replace default via <gateway> dev eth0 src <VEX IP> onlink table 172
ExecStartPre=+-/usr/sbin/ip rule del uidrange <vexillum uid>-<vexillum uid> priority 1000 table 172
ExecStartPre=+/usr/sbin/ip rule add uidrange <vexillum uid>-<vexillum uid> priority 1000 table 172

If your server has only one IP address, you do not need this.

5. First start and the admin page

  1. The first start needs a temporary admin password in the environment variable VEX_ADMIN_BOOTSTRAP_PASSWORD. It must be at least 12 characters. My first try with a short one stopped the service with an error. I put it in a root-only file loaded with EnvironmentFile=- (the dash means "fine if the file is missing").
  2. Start the service. Check http://127.0.0.1:8080/healthz on the server and the public liveness page on 8085.
  3. From your own computer, open a tunnel: ssh -N -L 8080:127.0.0.1:8080 root@your-server, then browse to http://localhost:8080. Log in as admin and change the password right away.
  4. Delete the bootstrap file. The service restarts fine without it.

A fresh install starts with no instances at all. Nothing is created for you. You add each mode yourself.

6. Add the modes (instances)

Each mode is an "instance", created on the admin page. Before each one: add its port to the service lockdown, open the firewall, restart. Then create the instance and click Start. Once my first click on Start did nothing, and a second click worked.

ModeWhat I enteredLessons
M17UDP 17002, designator JET, endpoint vex.n6jet.com:17002No address field: it listens on all addresses (see Trick 2). Modules A to Z.
DMRListen VEX IP:62031, talk TG 7, default module A, unlink 64000, status 65000Modules are reached as TG 64001 to 64026, then you talk on TG 7. The password is stored as plain text. VEXJET uses the open password passw0rd, like XLX and URF. I changed it in the database because the admin page had no field for it.
YSFListen VEX IP:42000, designator 95114, room name VEXJETOne room only, no DG-ID rooms. Use a 5-digit number as the designator, or it can never appear in hotspot host lists. There is no edit page for YSF instances in v0.2.0b, so changing the designator meant editing the database with the service stopped.
NXDNListen VEX IP:41400, TG 9512One room. Answers the standard NXDNGateway poll.
P25Listen 41100, TG 9512One room. Answers the standard P25Gateway poll.
D-STARDExtra VEX IP:30001, designator JET, modules A to ZOnly DExtra was offered in the form, not DPlus or DCS. A test link got an ACK right away.

Lessons

7. Put the API and dashboard on the web

Vexillum has no dashboard of its own. It has a read-only public API (/api/public/v1/...) that dashboards and DVRef read. I added W0CHP's free vex-dashboard, built from source with Go, listening on 127.0.0.1:8090.

The key is to serve both from one https address. My Apache site for vex.n6jet.com sends the API paths to Vexillum and everything else to the dashboard:

ProxyPreserveHost On
ProxyPass /api/public/ http://127.0.0.1:8085/api/public/ flushpackets=on
ProxyPassReverse /api/public/ http://127.0.0.1:8085/api/public/
ProxyPass / http://127.0.0.1:8090/
ProxyPassReverse / http://127.0.0.1:8090/

flushpackets=on keeps the live event streams flowing. A free Let's Encrypt certificate covers the https side. Now https://vex.n6jet.com/ shows the dashboard and https://vex.n6jet.com/api/public/v1/liveness answers for DVRef.

The most common mistake

If your Dashboard URL points straight at the dashboard program's port (for example http://yourhost:8090), the API is not there. The dashboard answers "404 page not found", and DVRef shows "Vexillum liveness API returned an error: HTTP 404". Another VEX operator ran into exactly this in November 2026. Put both behind one https address as shown above.

The dashboard can show an About tab from a Markdown file (about.md, set with an environment variable). To change it, edit the file and restart the dashboard.

8. List it on DVRef

DVRef is the directory that hotspot host lists are built from. A VEX server is listed in two layers: one VEX Server record, then one instance under it for each mode.

  1. Create the VEX Server record from your DVRef account: name (50 characters or less), country, sponsor, description, and the Dashboard URL. You can create it before the server is built. It stays hidden until it has an instance.
  2. Add each instance only after that mode is running. DVRef saves it only if it can read the liveness page at your dashboard address showing that mode running, and a real test packet to your host name and port is answered.
  3. For DMR, DVRef fills in port 62030. Type your real port if it is different.
  4. The instance name on DVRef does not have to match the instance name inside Vexillum. Mine are all "VEXJET" on DVRef.

Do not register instances twice

Lessons

9. List it on VexDV

VexDV (activity.vexdv.io) is the project's own live directory of VEX servers and stations. Joining is optional.

  1. Add to the config file:
    [activity.public]
    enabled = true
    display_name = "VEX-JET (N6JET, San Jose CA)"
  2. Restart, then request enrollment with the built-in tool. Run it from /opt/vexillum as the service user, because the database path is relative:
    cd /opt/vexillum && runuser -u vexillum -- ./vexillum publisher request -callsign YOURCALL -email you@example.com -config /opt/vexillum/vexillum.toml
  3. VexDV emails a code. Run publisher verify with it before it expires (24 hours).
  4. A VexDV admin approves the request. Mine was approved the same day. The certificate installs itself and renews automatically.
  5. On the admin page, open Publishing, find the Public Advertising card, and switch on each instance you want listed. I forgot NXDN and P25 at first, and only three of my six modes showed up.

Check progress any time with publisher status.

10. Hotspots and host files

Because VEX instances may not reach the hotspot host files, I add VEXJET to my hotspots by hand. On Pi-Star the custom host files live in /root/ (for example /root/DMR_Hosts.txt, /root/YSFHosts.txt, /root/M17Hosts.txt) and survive the nightly update. On WPSD use the Hosts File Editors. Step-by-step instructions with the exact lines are on Vexillum (VEX) Reflector Explained and How to Connect to N6JET Reflectors.

When you pick an alias for a custom host, search the full host lists first. My first D-STAR alias, XRFJET, turned out to be a real registered name (it belongs to XLXJET), and my gateway linked to the wrong reflector. I renamed it XRFVEX.

11. Bridges

Bridges are small separate programs from the Vexillum project that link a VEX room to another network or another mode. VEXJET runs three:

BridgeJoinsNotes
vex-dmr-bridgeTGIF TG 95110 and VEXJET DMR module ALogs into VEXJET like a repeater and into TGIF like a hotspot, with its own DMR ID and my TGIF hotspot password.
vex-dmr-ysf-bridgeDMR module A and YSF 95114Same voice codec on both sides, so no conversion. Uses the RadioID database to show callsigns.
vex-nxdn-ysf-bridgeNXDN 9512 and YSF 95114Reaches DMR through the YSF bridge.

Which bridge version to build

The newest bridge code uses a new "bridgelink" protocol with a token. Vexillum v0.2.0b does not support it: nothing listens on its port, and no token appears when you designate a bridge. The older versions from before bridgelink are still in the public code history, and they work with v0.2.0b. Build them from source with Go 1.25 or newer:

git clone https://repo.kc1awv.net/Vexillum/vex-dmr-bridge.git
cd vex-dmr-bridge
git checkout 28c1bf0
go build

The versions VEXJET uses: vex-dmr-bridge 28c1bf0, vex-dmr-ysf-bridge 05bb745, vex-nxdn-ysf-bridge d8f652c.

Lessons

12. Joining the modes (converter)

Vexillum has no bridge for M17, P25 or D-STAR. To put those on module A with DMR, I built a converter from DVSwitch parts, the same pattern as my older "N6JET Six Link". Each mode gets a "leg" that turns its digital voice into plain audio, and a small relay program passes the audio between the legs so each leg hears the others but never itself. The relay is explained on The Multimode USRP Fanout Bridge.

Lessons

13. D-STAR and the hardware vocoder

D-STAR uses an older AMBE codec that needs a hardware chip. I used two spare DVstick 30s plugged into a small Linux box at home, linked to the server over Tailscale. Four problems had to be solved:

  1. ircDDBGateway grabs port 30001 on every address, whatever its config says (I checked by watching what it opened). On a shared server that clashes with xlxd and urfd. Fix: run the gateway inside its own small network namespace, linked to the host by a virtual cable.
  2. The vocoder must sit next to the stick. With Analog_Bridge on the server and the stick at home, every voice frame waited a full round trip, about 80 ms, when it needed 20 ms. Fix: run Analog_Bridge on the home box beside the sticks. Only the audio stream crosses Tailscale.
  3. Analog_Bridge 1.6.4 only accepts a chip that reports "AMBE3000R". The DVstick 30 reports "AMBE3000F". A small local relay rewrites the answer.
  4. The sticks share a USB driver with the XLXJET transcoder sticks. The transcoder program unloads the serial driver when it starts, then reloads it so the driver takes only the two DVstick 30s.

After that, a 3-second test tone sent into the D-STAR leg came out as 156 D-STAR voice frames on VEXJET module A, in real time. The same worked on module B.

14. All the lessons in one list

  1. Check every port on every IP address before you pick yours. A program bound to 0.0.0.0 holds the port everywhere.
  2. Keep the admin, metrics and API listeners on 127.0.0.1. Use an SSH tunnel for the admin page.
  3. The bootstrap admin password must be at least 12 characters. Change it on first login, then delete the bootstrap file.
  4. A new install has no instances. Mode settings live in the database, not the config file.
  5. Run tools from /opt/vexillum, because the database path is relative.
  6. One port per SocketBindAllow line.
  7. M17 listens on all addresses. On a multi-address server, route by user so replies leave from the right address.
  8. Use 5-digit numbers for YSF so the room can reach host lists.
  9. Some settings have no edit page in v0.2.0b. Stop the service and back up the database before editing it by hand.
  10. Serve the dashboard and the API from one https address. A Dashboard URL that points at the dashboard's own port gives DVRef a 404.
  11. On DVRef, create the VEX Server record first, then add instances only when they are running.
  12. Never register a VEX instance number as a separate reflector. To convert an old listing, delete it first, then add it as an instance.
  13. VEX instances may not reach the hotspot host files. Add your server to hotspots by hand.
  14. Keep DVRef tickets plain: no IP addresses, lists or the word "delete".
  15. On VexDV, switch on Public Advertising for every instance.
  16. Use the pre-bridgelink bridge versions with v0.2.0b.
  17. Bridge descriptions: 19 bytes at most.
  18. Tie bridges and converter legs to Vexillum with PartOf= and a short start delay.
  19. Designate every bridge on the admin Bridges page.
  20. Search the full host lists before choosing any alias or number.
  21. D-STAR needs a hardware vocoder, and the vocoder program must run on the same machine as the stick.
  22. Test each step from outside the server, and write down every port, ID and file as you go.

15. Sources

Most of this page comes from one origin, my own build, checked against the project's documentation. Your server will be different. Test as you go.


A noncommercial hobby reference compiled by N6JET, gathered from hands-on work and public sources, and shared freely for anyone interested in amateur digital voice. Version 1.2, 11 November 2026.