The "Active WebUI" card (Settings > Web Applications) only showed
description and __APP_VERSION__, never the resourceUrl it's actually
built from — add it as a linked "Source" row.
That card also turned out to be misleading in local dev: it reads the
x:Application record whose urlPrefix matches /admin or /account, and a
freshly bootstrapped server keeps Stalwart's seeded default there
("Stalwart Web Interface", pointing at stalwartlabs/webui's release),
regardless of what's actually being served — in dev that's this fork,
served directly by Vite, which never touches that record at all.
dev-server-init.sh/.ps1 now point that record's description and
resourceUrl at this fork's own release, so the card (now including the
visible Source URL) reflects what's actually running instead of
Stalwart's factory default.
Verified end-to-end against a fresh container with both scripts.
6.4 KiB
Development
How to run a local Stalwart test server and develop this WebUI against it. This file is meant to be read by humans and AI coding agents alike — see AGENTS.md for the rules that also apply while doing this.
Prerequisites
- Node.js 18+
- Docker (with the
docker composeCLI plugin)
1. Start a local test server
npm run dev:server
This runs docker compose up -d, which starts a disposable Stalwart
instance (docker-compose.yml) with:
- The management/JMAP HTTP API on
http://localhost:8080(the only port the WebUI dev proxy needs — seeserver.proxyinvite.config.ts). - Mail protocol ports (SMTP/IMAP/POP3/ManageSieve) exposed too, only needed if you're testing actual mail flows, not just admin UI screens.
- A fixed admin account baked in via
STALWART_RECOVERY_ADMIN:admin@example.org/c8321iEscHDy0GWV. Disposable dev credentials only — never reuse them for anything real. - Named Docker volumes (
stalwart-etc,stalwart-data) so data survives a restart. Data persists until you tear the volumes down.
Useful companions:
npm run dev:server:logs # tail the container's logs
npm run dev:server:down # stop it (add `-v` via `docker compose down -v` to also wipe data)
The server takes a couple of seconds to come up; docker compose logs stalwart
will show Network listener started ... localPort = 8080 once it's ready.
2. Initialize the server (first time only)
The container starts in Stalwart's bootstrap mode, which only allows
signing in as the break-glass STALWART_RECOVERY_ADMIN account — real
accounts and most settings aren't usable yet. Run once per fresh volume:
# Windows / PowerShell
pwsh ./scripts/dev-server-init.ps1
# Linux / macOS / any POSIX shell (including most AI agent sandboxes)
bash ./scripts/dev-server-init.sh
This completes the bootstrap wizard (default domain example.org, no TLS
certificate request — safe for local/offline use), creates a real
devadmin@example.org admin account, sets the server's default OAuth
access token lifetime to 3 hours, and points the x:Application record
serving /admin//account at this fork's description and release URL
(cosmetic — otherwise Settings > Web Applications' "Active WebUI" card
shows Stalwart's default "Stalwart Web Interface" / upstream release URL,
even though what's actually running is this fork served by Vite). It's
idempotent — safe to re-run, it no-ops once the server is already
bootstrapped. You only need to re-run it after docker compose down -v
(which wipes the volumes).
3. Get an access token
For local development it's simpler to skip interactive login and use a
bearer token directly via VITE_ACCESS_TOKEN (see .env.development).
# Windows / PowerShell
pwsh ./scripts/dev-token.ps1 # 3 hour token (server default)
pwsh ./scripts/dev-token.ps1 -DurationSeconds 1800 # custom duration (30 min)
# Linux / macOS / any POSIX shell (including most AI agent sandboxes)
bash ./scripts/dev-token.sh # 3 hour token
bash ./scripts/dev-token.sh 1800 # custom duration (30 min)
Both scripts authenticate as the devadmin account created in step 2 and
create a Stalwart API key with the requested expiry (default 3 hours,
overridable per invocation — this is a genuine per-request duration, not
a global setting), then write its secret to .env.development.local
(gitignored, never committed) as VITE_ACCESS_TOKEN. Re-run the script
and restart npm run dev once the token expires (the UI starts returning
401s).
The STALWART_RECOVERY_ADMIN account is intentionally not used here: it's
a break-glass credential and its tokens always expire in a fixed 1 hour
regardless of server configuration, so it can't honor a custom duration.
4. Run the WebUI
npm install # first time only
npm run dev
Open http://localhost:5173. You should land directly in the admin panel
(no login screen) since VITE_ACCESS_TOKEN is set.
5. Verify your change
npm run typecheck
npm run lint
npm test
npm run build
For UI changes, actually look at the running app (browser or a browser automation tool) — passing typecheck/lint/tests proves the code compiles and existing behavior didn't regress, it doesn't prove the new UI works.
Resetting the test server
To start from a completely clean server (e.g. to re-test first-run behavior):
npm run dev:server:down
docker compose down -v # also removes the stalwart-etc/stalwart-data volumes
npm run dev:server
bash ./scripts/dev-server-init.sh # re-run: fresh volume needs bootstrapping again
Troubleshooting
- UI shows a login screen instead of the admin panel, or the page fails
to load data:
npm run devonly reads.env.development.localat startup. If you regenerate a token, or restart the Docker container, whilenpm run devis already running, stop it (Ctrl+C) and start it again — it won't pick up the new token or reconnect on its own. Failed to fetch/ connection refused in the browser console: the test server isn't running or isn't ready yet. Check withdocker ps --filter name=stalwart-webui-devandnpm run dev:server:logs; wait forNetwork listener started ... localPort = 8080before retrying.- 401s after everything was working: your token expired. Re-run
scripts/dev-token.sh(or.ps1) and restartnpm run dev. scripts/dev-server-init.shfails with connection errors: the container needs a few seconds afternpm run dev:serverbefore it accepts requests — the script retries for ~30s, but if your machine is slow, just re-run it.
Notes for AI agents
- This whole workflow (steps 1–4) is scriptable end-to-end without a
browser:
npm run dev:server, thenbash scripts/dev-server-init.sh(first time only), thenbash scripts/dev-token.sh, then the app is reachable athttp://localhost:5173withVITE_ACCESS_TOKENalready set. Verify backend connectivity directly withcurl, e.g.curl -H "Authorization: Bearer $TOKEN" http://localhost:8080/jmap/session. - Read AGENTS.md before touching anything under
src/— the schema-fidelity rule applies to all development, local test server or not. - Don't commit
.env.development.local(it holds a live token) or leave the dev container running unexpectedly —npm run dev:server:downwhen you're done.