WISP32 User Manual

Three things to do before anything else: turn it on, connect it to your network, and find it again from a browser.

Turning it on

Press and hold BOOT until the display turns on. The device boots to the clock screen — this is home base. Every other page is reached from here by swiping.

BOOT (top) and PWR (bottom) on the case edge.

Connecting to WiFi

On a brand-new device (or after a reset), it can't reach the internet yet, so it opens its own WiFi network so you can configure it:

1. On your phone or computer, connect to the WiFi network Watch-Setup. This can take a few minutes to appear and connect — and while you're on it, your phone or computer has no internet access, since it's a temporary offline network broadcast by the device itself.
   Watch-Setup
2. A setup page should open automatically — if not, open a browser and go to the address shown on the device's screen.
   WiFi Setup

   1. Connect to WiFi network:

   Watch-Setup

   2. Open a browser to:

   192.168.4.1

   Swipe down for Settings

   What the device shows during setup

3. Enter your home WiFi network's name and password, then click Save.

   WiFi Networks

   The device tries each of these in order at boot. The first one is also treated as "home" for the Auto Sleep override on the device's Settings page 6.

   My Network

   ••••••••

   + Add another network

   Tip: if the device has no working WiFi at all, use Settings page 6's "WiFi Setup" row on the device itself instead — this page needs an existing connection to reach it.

4. The device reboots and connects to your network.

You can add additional networks later from the Credentials page in the web server — useful if you move the device between locations, like home and office. It also works well paired with your phone's personal hotspot when you're away from a saved network — the watch is built to complement your phone, not replace it, so it works anywhere your phone has signal. Keep in mind a weak signal slows things down either way: expect slower responses and reduced performance the further you are from whatever access point you're connected to.

Finding your device on the network

Once connected, the device is reachable at:

http://<device name>.local

Every device starts out with a generated name like watch-a4cf1 (from its own hardware ID, so it's already unique — see Device Name below), making the default address something like http://watch-a4cf1.local. If your network doesn't support that address (some routers and phones don't), check the clock screen — it shows the device's current identifier, which you can type into a browser instead, or switch the clock face to show the IP address directly (e.g. http://192.168.1.42) — see Device ID on the web server's Configuration page.

Device Name

Set from the Credentials page, at the top. This is what the device calls itself on the network (its .local address) and, by default, what the clock face shows to identify it. Only letters, numbers, and hyphens are kept from what you type — spaces become hyphens, everything else is dropped; leave it blank to go back to the generated default.

Saving a changed name restarts the device to apply it, same as a WiFi network change.

First connection to the web server

The first time you open the web server, expect to see one or more warning banners above the nav tabs. They're not errors — they're a running checklist of setup gaps, re-checked on every page load. Each one disappears on its own the moment it's fixed. Red means the gap leaves something exposed; orange means a feature just won't work yet.

Basic navigation

• Swipe left, right, up, or down from the clock screen to reach the AI Assistant, Translator, Voice Notes, and other pages.
• Tap BOOT briefly to turn the display on or off.
• Hold BOOT down for a few seconds to power the device off — a countdown appears on screen; release early to cancel.

Hold a button, ask a question, get a spoken answer back — powered by Gemini.

Setting up your API key

The Assistant uses Gemini (Google). You need an API key:

1. Open the web server and go to the Credentials page.
2. Paste in your Gemini API key.
3. Save.

On the device, the Model setting shows "No Key" until you've added a key.

Talking to it

1. Swipe left from the clock to open the AI Assistant page.
2. Hold down BOOT while you speak.
3. Release BOOT when you're done — the Assistant thinks, then replies.
4. If it's talking or thinking and you want to interrupt, just tap BOOT once — this stops it immediately ("barge-in") instead of starting a new recording.

Display modes

Tap the Mode row to cycle through:

• Voice — audio only, nothing shown on screen.
• Dual — audio plays and the conversation text is also shown.

Replaying the last answer

Once the Assistant has finished a reply, swipe down anywhere on the page to hear it again — no new request is sent. A "Swipe down: replay" hint appears on screen whenever this is available. Replay holds up to 30 seconds of audio and plays straight from memory, so it works whether or not Save History is turned on. A reply longer than that cuts off near the end, and starting a new recording clears whatever was there to replay.

Reviewing tool calls

Swipe up to see a short history of what the Assistant has actually done during this session (alarms set, files opened, searches run). Swipe down to close it.

What it can do

Beyond conversation, the Assistant can act on your behalf:

• Set, list, and cancel alarms and timers by name.
• Search the web for current information.
• Answer questions about files in the Knowledge Base.
• Take extra time to reason through a harder question ("deep think").
• Adjust the speaker volume.
• Send an email to a saved contact.
• Run one of your scheduled tasks by name.

Assistant settings

Reachable from the on-device Settings pages or the web server's AI Models page:

• Voice — which voice it speaks with.
• Save History — whether conversations get written to a browsable chat log.
• Reasoning/Thinking Effort — how much the model "thinks" before answering.
• Max Output Tokens — a cap on how long a single reply can run.
• System Prompt — customize the base instructions the Assistant follows, from the web server's AI Models page.

Why the screen sometimes goes dark for a moment after closing this page

A short voice session uses a large amount of the device's working memory for audio processing. To keep the device reliable over many sessions, closing AI Assistant, Translator, or Voice Notes silently triggers a quick memory-clearing cycle a moment later — the screen stays off throughout, so from the outside it just looks like the display briefly not responding before returning to normal. This is expected, not a fault, and doesn't happen if Auto Sleep (Settings 6/6) is already putting the device to sleep on its own between uses.

Real-time, two-way spoken translation, using the same AI connection as the Assistant.

1. Swipe right from the clock to open the Translator page.
2. The Source and Target rows show the two languages in use — tap either to step through your language list, or use Switch to swap direction.
3. Hold BOOT to speak in the source language; release to hear it translated aloud in the target language.

Swipe down after a translation finishes to hear it again, same as the AI Assistant page.

Choosing which languages appear

By default only a curated subset appears on the device, so you're not scrolling through hundreds of entries. To change it:

1. Open the Languages page in the web server.
2. Move languages between the "Available" and "Included" lists — the order you build the Included list in is the order they'll cycle through on the device.
3. Save.

Like AI Assistant, closing this page may cause a brief, silent memory-clearing cycle a moment later — see the note in the AI Assistant section for what's happening and why it's expected.

A quick way to capture a spoken thought as a saved, transcribed note. Recording length is not capped — tap to start, talk for as long as you want, tap to stop.

1. From the clock screen, swipe up from the bottom edge to open Voice Notes.
2. Tap the middle of the screen to start recording; tap again to stop, whenever you're done.
3. The note is automatically transcribed, given a title, and saved into the Knowledge Base — no extra confirmation step. Most of the transcription happens continuously in the background while you're still talking, so only a short wait remains after you tap stop.
4. Swipe down to close and return to the clock.

Under the hood, a recording is split into short chunks as it's captured, each one transcribed and saved the moment it's ready rather than waiting for the whole recording to finish. If a chunk fails to transcribe (a network hiccup, for example), the rest of the note isn't affected — everything transcribed before and after that point is still saved; only that one piece's text is missing.

Two optional rows on this page control what happens after a note is saved:

• Save Audio — a safety net, not just a copy. When on, the device keeps a continuous backup recording on the SD card as the note is captured. If a chunk's transcription ever fails, its audio is still there and recoverable, even though that piece's text is lost. When off, a failed chunk's audio is not kept either — only text that was successfully transcribed is saved.
• Email To — automatically email the transcript to a saved contact as soon as the note is saved. Cycle through your saved contacts to pick one, or leave it off.

Like AI Assistant, closing this page may cause a brief, silent memory-clearing cycle a moment later — see the note in the AI Assistant section for what's happening and why it's expected.

Show a QR code full-screen on the device — handy for sharing WiFi credentials, a URL, or any short text without anyone needing to type it in.

1. Press and hold the clock face (without moving your finger) for about 2 seconds.
2. The first code shown is always this device's own web dashboard address, so it's reachable even before you've saved anything.
3. Tap PWR to step through any codes you've saved, one at a time, wrapping back to the dashboard address after the last one.
4. Tap BOOT to close and return to the clock. The page also closes on its own after the usual screen-sleep timeout.

A label too long to fit scrolls across the screen automatically, stopping after 3 passes rather than looping forever.

Saving your own QR codes

Managed entirely from the web server's Configuration page, under the QR Codes card:

1. Fill in a Label (up to 50 characters) and the Content to encode (up to 500 characters — multi-line text is fine).
2. Tap Add QR code.

Up to 25 can be saved; once you hit that limit, delete one to make room for another. Each saved entry shows a short preview — tap it to see the full label and content, use the up/down arrows to reorder the list (the on-device PWR cycle follows this same order), or Delete to remove it.

The easiest way to set one is simply to ask the AI Assistant by voice.

• "Set a timer for 10 minutes called Pasta."
• "Set an alarm for 7am called Wake Up."
• "Cancel the Wake Up alarm."
• "What alarms do I have?"

Alarms and timers can be named, which makes it easy to have several going at once and cancel the right one. When one fires, its name appears on screen.

Have the Assistant carry out a set of instructions unattended — once, on a schedule, or whenever you ask it to by voice. Managed entirely from the web server's Tasks page.

Creating a task

Fields on the Create Task form:

AI instruction vs. Action steps

Every step's card has a toggle at the top choosing which kind it is:

• AI instruction — a plain-language instruction sent to the task's Model. Grayed out and unavailable whenever the task's Model is None, since there'd be nothing to run it with — any step already set this way switches to Action automatically the moment Model is changed to None.
• Action — a direct call to one built-in capability (see the table below), with its own fields, no AI model involved at all. Faster and more predictable than describing the same thing in words, and works even with Model set to None.

Action steps

Sixteen built-in actions, each with its own fields. Several fields are dropdowns instead of free text — picking from what's actually there avoids typing a name or unit that can't possibly match anything:

Chaining steps together

Any step can be given a Save result as name. A later step can then reference that result by typing {{name}} into any of its own text fields — or click the {} button next to a field to pick from what's available instead of typing it by hand. Referencing a name that hasn't produced anything yet (a typo, or a step later in the list) leaves the literal {{name}} text in place rather than silently disappearing, so a misconfigured chain is visible instead of hidden.

What actually gets saved is each step's own full result text — e.g. a Calculate step saves "2+2 = 4.000000", not just 4. That's useful for dropping straight into an email body, less useful as a clean value to feed into another field expecting just a number or a name.

API Call steps

Instead of (or alongside) other steps, a task can have one fixed API Call configured — method, URL, headers, an authorization type (bearer token, basic auth, or an API-key header) and its value, and a request body. Use the Test button to fire it once and see the real response; click any value in that response to save it under a name, which then becomes available to every step via {{name}} the same way a step's own Save result as does.

Running a task

Tasks with an active schedule fire automatically when due, as long as the device is idle (screen off but awake, or has reconnected to WiFi after a deep sleep). You can also run any task immediately with its Run chip on the Tasks page, or by asking the Assistant by voice. Each task's card shows when it last ran and whether it succeeded, and can be moved up or down the list with the ▲/▼ buttons at its bottom-left — display order only, doesn't affect when or whether it actually runs.

Two separate storage areas, reached from different pages in the web server.

Knowledge Base

Where notes, saved documents, and anything the AI Assistant is asked to remember get stored. Open the web server's Knowledge page to browse folders and files — from there you can view, edit, delete, move, or download anything (via the ⋮ menu on each file), and create new folders to organize things. The Assistant can search and reference this content when you ask it questions.

File size and type limits

Uploads are capped at 500 KB, shown right above the file list along with the current list of accepted extensions — tap the small (i) button next to it for the reasoning. In short: the Assistant sends a file's entire content on every question about it, and most AI models cap how much text they can read at once around 128,000 tokens; 500 KB of plain text fits comfortably inside that with room to spare. Only plain-text formats are accepted (documents, notes, code, data files, calendar and email exports) — not Word, PDF, or other formats that pack in binary formatting data the Assistant can't actually read.

Protecting a file with a PIN

Sensitive files can be locked behind a numeric PIN (4–8 digits):

1. Find the file — or folder; protecting a folder protects everything inside it — in the web file browser.
2. Choose Protect and set a PIN.
3. From then on, opening that file — from the web server or by asking the AI Assistant about it — requires the PIN.

Changing or removing an existing PIN requires the current one; there's no bypass through the web login alone.

Files

A separate, simpler upload/download area (the Files page) for things like the audio recordings from Voice Notes or anything you've uploaded directly. It supports browsing, renaming, moving, and per-file encryption, but not PIN protection — that's specific to the Knowledge Base.

Two layers, one automatic and always on, one optional and stronger.

Automatic baseline protection

Your saved API keys and WiFi passwords, your task list (including anything typed into an API Call step's auth field), and every Knowledge Base PIN are automatically scrambled using a key derived from your SD card's own unique hardware ID. This happens silently — no setup, no password, no prompt.

What this protects against: someone casually pulling the card and browsing its files on a computer. What it does not protect against: someone who deliberately wants in — the key comes from the card itself, so anyone who has the physical card can, in principle, derive it too.

Full password encryption

For real protection — the kind that holds up even against someone who has the card — activate full encryption from Credentials → Encryption in the web server:

1. Open the Encryption page.
2. Choose whether this is the first time encryption has been activated on this card.
3. Set a password (at least 8 characters, with an uppercase letter, a lowercase letter, a number, and a special character).
4. Confirm — this encrypts every existing file with a key derived from your password combined with the card's ID.

Once active, this covers the entire Knowledge Base, chat history, prompts, and credentials — not just the baseline items above.

Device Lock (child lock)

A separate thing from the two encryption layers above — this doesn't touch your data at all, it's a physical-access toggle on the device's own screen. Turn it on (Settings 1/6 → Device Lock) and swiping into AI Assistant, Translator, or Voice Notes is blocked; the clock face and Settings stay fully reachable either way. What exactly happens on that swipe depends on whether a PIN is set:

• No PIN set — a brief on-screen notice, shown for 5 seconds, then back to the clock.
• PIN set — the same numeric keypad used to turn the lock off (below) opens directly instead of the notice. Entering the correct PIN opens that one page for this visit only; the lock itself stays on, so the very next swipe asks for the PIN again the same way. This is a quick way in without leaving the lock off for anyone else who picks up the device afterward.

Shown for 5 seconds after a swipe into a locked page (no PIN set), then back to the clock.

Turning it back on is always a plain tap. Turning it back off depends on whether a PIN is set:

• No PIN set (the default) — tapping Device Lock again just turns it off, same as any other Settings toggle.
• PIN set — tapping it instead opens a numeric keypad on the device. Enter the correct PIN to turn it off; a wrong entry clears and lets you try again, up to 3 times. On the 3rd miss, the device is bounced all the way back to the clock (still locked) and starts a 30-second lockout. Swipes don't work on this keypad screen — PWR is the only way to back out without entering it.

Reopening Settings and tapping Device Lock again during that 30 seconds shows the same keypad, but grayed out and unresponsive, with a live countdown ("Try again in 12s"…) where the retry count normally shows — it isn't a fresh set of 3 tries, and there's no way to skip the wait. Once the countdown reaches zero the keypad wakes back up on its own, no need to close and reopen it.

Set or change the PIN from Credentials → Device Lock PIN in the web server — 4 to 6 digits, or leave it blank to go back to no-PIN behavior. Forgot the PIN? Just clear or change it there; the web Credentials page is never itself behind this lock.

Reprogramming the firmware

Minimal. This device ships fully open — the firmware can be reflashed over USB at any time, by anyone with physical access to it. A lost or stolen device reveals nothing on its own: the encryption key is never stored on it (see Full password encryption, above), so it can't be decrypted.

The attack requires physical access to the device, reprogramming it with lookalike firmware, then waiting for you to re-enter your password — and requires gaining access to the device again, physical or over WiFi, to retrieve any information. This is a general tradeoff of open, reprogrammable hardware, not something unique to this device; a hardware lock (secure boot) is the only way to close it off, and this device doesn't ship with one. Modifying, extending, or replacing the firmware yourself is entirely your own responsibility.

Reach any of these by browsing to the device's address — http://<device name>.local (see Device Name in Getting Started) or its IP.

Six pages, reached by swiping down from the top of the clock screen (opens page 1), then a brief press of PWR to step through the rest: Clock → 1 → 2 → 3 → 4 → 5 → 6 → back to Clock. Swipe up on any page to close Settings immediately and return to the clock. Every value below is tap-to-cycle unless noted otherwise, and takes effect the moment you tap it — nothing needs a separate save.

Settings 1/6 — Volume, Display & Lock

Settings 2/6 — AI Assistant

Settings 3/6 — File Q&A & Deep Think

Settings 4/6 — Feature Models

Settings 5/6 — Output Limits & Voice

Settings 6/6 — Power & Network

Two physical controls, and both change behavior depending on what's on screen.

BOOT

PWR (side button)

Battery life

The watch has power management with auto deep-sleep when idle, so with optimal deep-sleep cycling and not constant usage it can last up to 12 hours.

Device won't connect to WiFi

Re-enter setup mode and reconfigure, or check/update your network's password on the Credentials web page. Make sure the device is within range of your router.

Can't reach the device by its .local address

Some networks don't support that kind of address at all. Check the clock screen for the device's current IP address and use that instead — or switch what the clock face shows via Device ID on the web server's Configuration page.

If you have two of these devices on the same network and only one is ever reachable by name, they likely share the same Device Name (see Device Name in Getting Started) — give each a different one from the Credentials page.

No sound

Check the volume slider — the low end of its range is quiet by design, so try somewhere around the middle before assuming something's wrong. If it still looks fine but there's no sound, try a full power cycle (power off completely, then back on); this clears up occasional audio hardware hiccups.

Swiping to AI Assistant, Translator, or Voice Notes does nothing at all

The current model for that page has no API key configured — the page doesn't open, and nothing is shown, on purpose (see the callout in the AI Assistant section). Check Credentials and add your Gemini API key.

AI Assistant won't respond, or shows an error

This is different from the above — a key is configured but the connection still failed. The screen shows a short error message and, underneath it, a more specific reason, most often the connection being closed by Gemini's own server — usually resolved by just trying again (see the connection-error shortcut in the AI Assistant section).

Voice Notes won't open ("Voice TT is off")

Settings 4/6's Voice TT row is set to "None." A note can't be saved without a transcript, so this is a deliberate block rather than a failed recording — pick a real model on that row to record notes again. If Voice TT already has a real model selected and the page still won't open, your Gemini API key is likely missing — see the entry above.

A task didn't run when scheduled

Scheduled tasks only fire while the device is idle. If the screen was actively in use, or two tasks were scheduled for the same moment, a run can get skipped — check the task's last-run status on the Tasks page, and see the caution note in the Task Scheduler section.

AI Assistant, Translator, and Voice Notes won't open after swapping SD cards

If the SD card now in the device has its own web server password (set on a different device, or previously on this one), the device locks itself to the clock face — swipes to every other page do nothing. Open the web server from a browser and log in with that card's password; a correct login is what unlocks the device and activates that card's API keys for on-device use. If the card doesn't have its own password, this doesn't apply — everything works immediately.

The clock face while this is happening — time still shows normally, only navigation is blocked.

Forgot the Device Lock PIN

Not a lockout — the web Credentials page is never itself behind Device Lock. Open Credentials → Device Lock PIN in a browser and clear the field (or set a new one) and Save; the device picks it up immediately, no restart needed. See Device Lock in the Security section for the full behavior.

"Insert SD Card" appears with the card still in, or takes a moment to notice a card is gone

This device has no dedicated card-detect switch wired up, so it can't sense the card electrically the instant it's removed or inserted — it has to actually try reading the card every few seconds and notice when that stops working. A newly inserted card can take a couple of seconds to be confirmed before "SD Card Found" appears, and a removed card can likewise take a few seconds to be caught. This is expected behavior, not a fault — if the screen settles on the correct state within about 5–10 seconds either way, nothing is wrong.

Audio Player (web Apps tab) won't play anything

Check whether AI Assistant or Translator is open on the device — the Audio Player page shows a banner explaining this the moment it applies. Both features need the same limited working memory a live voice session already holds; close the device page and try again.

A Knowledge Base file won't upload, or the Assistant says a file is "too large"

Files are capped at 500 KB and limited to plain-text formats — see File size and type limits in the Knowledge Base & Files section for the full list and the reasoning behind it. A file over the limit needs to be split into smaller pieces, or trimmed down, before it can be uploaded or opened.

Settings at a glance