XIAO ESP32S3 camera web apps — STA (router) mode

The board joins the user's WiFi router. The PC/phone stays on its normal network (internet keeps working) and reaches the board at http://<mdns-name>.local or its DHCP IP.

Read the xiao-esp32s3 skill (same repo) first for base workflow and pitfalls (especially: PSRAM flag mandatory, DTR reset-stuck recovery, no serial reads while a demo must keep running).

Step 1 — ALWAYS ask the user first

Never invent, guess, or silently reuse WiFi credentials. Ask the user for:

1. 공유기 SSID — must be a 2.4 GHz network (ESP32 cannot see 5 GHz;

if the user gives something_5G, ask for the 2.4 GHz band name).

1. 공유기 비밀번호.
2. mDNS 이름 (optional, default xiao) — page becomes

http://<name>.local. Boards on the same LAN must each get a UNIQUE name.

1. Which app: collector, inference viewer, or both.

Step 2 — generate the sketch from the template

Templates in assets/ are verified working code — do not rewrite them:

• assets/web_collect.ino.tpl — dataset collector (gallery + delete + ZIP download)
• assets/web_infer.ino.tpl — live inference viewer (needs the

<project>_inferencing Edge Impulse library installed in the sketchbook)

Copy the template to <workspace>\<NN><name>\<NN><name>.ino (folder = ino name), then replace the placeholders literally:

The template already falls back to a XIAO_CAM hotspot if the router is unreachable for 15 s, so a wrong password degrades gracefully.

Step 3 — compile, upload

arduino-cli compile --fqbn esp32:esp32:XIAO_ESP32S3 --board-options PSRAM=opi <SKETCH_DIR>
arduino-cli upload -p COM4 --fqbn esp32:esp32:XIAO_ESP32S3 --board-options PSRAM=opi <SKETCH_DIR>

PSRAM=opi is mandatory (camera + model need PSRAM). The inference build needs --clean if the Edge Impulse library was just swapped.

Step 4 — verify end-to-end

The PC is on the same LAN, so HTTP is the fastest proof the whole path works:

Start-Sleep -Seconds 20   # boot + WiFi join
curl.exe -s -m 15 http://<mdns-name>.local/ -o "$env:TEMP\page.html"
(Get-Item "$env:TEMP\page.html").Length     # > 1000 bytes = page served
# inference app only:
curl.exe -s -m 15 http://<mdns-name>.local/classify   # JSON with scores

If mDNS resolution is flaky, find the IP once (ping <name>.local or router DHCP table) and use it directly — report BOTH addresses to the user.

Watching the board over serial

HTTP proves the page is up, but it cannot show you why a board failed to join the router. Serial can: the sketch prints the SSID it tried, the IP it got, and the mDNS name. Use the xiao-serial-monitor skill for it.

mon                      # interactive, port auto-detected
.\scripts\mon.ps1 -Seconds 15    # bounded, for your own debugging

Expected on success:

Connecting to WiFi ... IP: 192.168.0.179
mDNS: http://<mdns-name>.local
Collector ready

Two things worth knowing rather than avoiding:

• Opening the port resets the board via DTR, so the log restarts from boot.

That is usually what you want here — the WiFi join messages only print once.

• If a board ends up stuck showing only the ROM banner afterwards

(pitfall 8 of xiao-esp32s3), re-running arduino-cli upload clears it.

When the user wants to watch it themselves, hand them the command rather than launching the interactive monitor from a tool call — it never returns. In Claude Code the ! prefix runs it in their session:

! <path>\mon.ps1

Notes

• Collector page: captures go to an in-browser gallery (label badges, per-shot

delete, "전체 ZIP 다운로드" produces label.N.jpg files ready for Edge Impulse). Refreshing the page clears the gallery — download the ZIP first.

• Inference page: whole-frame colored box + top label + per-class confidence

bars (classification model — per-object location boxes need a FOMO model).

• The camera renders rotated 90°; the inference template already compensates

(CW feature rotation, verified empirically).