Huginn – Support
Welcome to Huginn — Gmail triage with a language model that runs entirely on your iPhone. Here you'll find answers to common questions and ways to get help.
Frequently Asked Questions
What is Huginn?
Huginn connects to your Gmail, reads your unread mail, and files it into categories you define in plain English. Instead of building filter rules, you describe what belongs in a category — "receipts I need for expenses", "things that need a reply this week" — and an on-device model does the sorting. Each category can then run actions automatically: archive, label, notify, extract fields, or add a reminder.
Do I need a Google account?
Yes. Huginn works with one Gmail account and signs you in through Google's own
web sheet using OAuth 2.0. It requests a single permission, gmail.modify,
which lets it read your mail and change labels. It cannot send email as you, and
it never sees your Google password.
Is my email sent to a server?
No. Classification, summarization and field extraction all run on your iPhone, on Apple Foundation Models or on Qwen3 through MLX. Huginn has no backend — the only servers it contacts are Google's, and Hugging Face once if you download the Qwen weights.
Why isn't it classifying anything?
Because it isn't supposed to yet. Huginn starts read-only: it loads your unread mail into the Review tab and waits. Emails without a category show as "Queued" — that's the normal starting state, not an error.
Open any email and tap Analyze to run the model on it. When you're ready for it to work on its own, turn on Classify automatically in Settings, and Run actions when you want it to start changing your mailbox. Those two switches are off by default on purpose: automatic bulk changes to a real Gmail account are worth opting into deliberately.
Which model does it use, and do I have to download anything?
Two backends, chosen in Settings:
- Apple Foundation Models — the default on devices with Apple Intelligence available and enabled. Nothing to download.
- Qwen3 1.7B (4-bit) via MLX — works on any supported device, and downloads about 1.1 GB of weights once, over Wi-Fi.
If your chosen backend isn't ready — Apple Intelligence turned off, or the Qwen weights not downloaded yet — Huginn pauses and tells you what to do instead of failing silently.
Why does analysis take a few seconds?
Because it's running a language model on your phone rather than calling a server. The Qwen backend is a reasoning model and thinks before it answers, which is slower but more accurate on messy marketing HTML. It's also fully deterministic — the same email always produces the same answer.
It put an email in the wrong category. Now what?
Assign the right one. That correction is the training signal: Huginn remembers the structural pattern of that kind of message, how you've treated that sender before, and the decision itself as an example for future prompts. Similar mail gets filed instantly afterwards without running the model at all — so correcting it makes the app both more accurate and faster.
What can actions actually do?
Per category, in order: mark as read, archive, star, apply a Gmail label, send a notification, log to the in-app feed, extract structured fields from the body, save a link for later, add an item to Reminders, or fire a notification that deep-links into another app. Actions are idempotent — if a run is interrupted, resuming it won't archive or notify twice.
Troubleshooting
-
Gmail won't connect, or sign-in fails Make sure you're signing in with a valid Google account and that Huginn still has permission. You can check and revoke access under Security → Third-party apps in your Google Account. Reconnecting from the app re-issues the token.
-
"Apple Intelligence isn't available" Either enable Apple Intelligence in iOS Settings, or switch the backend to Qwen3 in Huginn's Settings and download the weights.
-
The model download stalls or fails It's a ~1.1 GB download — stay on Wi-Fi and keep the app in the foreground. If it fails partway, retry from Settings; it resumes rather than starting over.
-
The app was killed while analyzing Model weights are memory-hungry, and iOS will reclaim memory aggressively on older devices. Huginn unloads the model when it backgrounds and after a short idle. If it keeps happening, close other apps, or use the Apple Foundation Models backend, which has no such footprint.
-
Actions didn't run Check that Run actions is on in Settings, that the category has action steps configured, and — for notification or reminder steps — that you granted the relevant iOS permission when prompted.
-
Crashes or unexpected behaviour Make sure you're on the latest version, then get in touch with your device model and iOS version.
Feedback & Feature Requests
Huginn is actively developed and feedback genuinely shapes what gets built next — especially reports of email that gets consistently misfiled, which is the most useful thing you can send.
Contact Support
If you need help or have questions, contact:
We aim to respond as quickly as possible.
Thank you for using Huginn.