Description

After a clean install (git clone + bun install + gbrain init --pglite), running gbrain serve --http --port 8787 --bind 0.0.0.0 results in /admin returning Cannot GET /admin (404).

This happens even when accessing directly via localhost (no reverse proxy involved).

Steps to Reproduce

1. Fresh clone of gbrain on Raspberry Pi 5 (Debian)
2. bun install
3. bunx gbrain init --pglite (chose conservative mode)
4. Start with: gbrain serve --http --port 8787 --bind 0.0.0.0
5. Access http://localhost:8787/admin or http://127.0.0.1:8787/admin

Expected Behavior

The admin dashboard should be accessible (as advertised in the startup banner: Admin: http://localhost:8787/admin).

Actual Behavior

Returns:
Cannot GET /admin
text(HTTP 404 from Express)

Environment

• GBrain version: 0.35.1.0
• Engine: PGLite (fresh database)
• OS: Debian (Raspberry Pi 5)
• Run method: systemd service + manual start both tested
• Also tested via Caddy reverse proxy on public subdomain (same result)

Additional Context

• gbrain doctor reports healthy (95/100).
• /health endpoint works correctly.
• The startup banner always prints the admin URL and a token, but the route does not respond.
• This behavior started appearing after PGLite locking issues in a previous install. A full reinstall + fresh gbrain init did not resolve it.

Questions for Maintainers

• Is the web admin dashboard expected to work out-of-the-box after gbrain init + serve --http?
• Does it require additional initialization steps (e.g. OAuth client creation, bootstrap token usage, or specific environment variables)?
• Is the admin SPA only served under certain conditions?

Happy to provide more logs or reproduction steps if needed.