hawmie

Privacy policy

Last updated October 13, 2026

hawmie is a private family archive: home movies, recordings, photos and other files, kept for a family and shared only with the people they invite. Jon runs it. This page says what we keep, which companies help us run it and what each of them sees, and how to delete things.

Short version: your files are private, nobody outside the library can see them, nothing is sold or used for advertising, and the optional AI features only run when the library's owner, or an admin they chose, starts them.

What we keep

We use two cookies: one that keeps you signed in, and one that remembers whether you like the list or the grid. There are no ads, no analytics and no tracking cookies.

Who can see your files

Only people the library's owner invites, and only the folders they were given. Files are never public: every picture, video and download goes through a link that expires within hours, made only after we check who's asking. Search engines are told not to index hawmie.

Companies that help run hawmie

These services store or process data for us. None of them may use it for anything but running hawmie.

The AI features

“AI features” are three optional tools, shown together as Smart features, that use machine-learning models to read what's in a video or recording and fill in things you'd otherwise type by hand: what was said, who appears, and when it was filmed. Nothing is analyzed when you upload. They only run when the library's owner, or an admin they chose for a folder, presses Smart features on a folder, a file or a selection and picks which ones to run. Making videos playable (converting formats, posters, waveforms) isn't AI and runs on our own systems.

All three run in our background jobs on Trigger.dev, which reads the file from our storage, does the work, and deletes its working copy when the job ends. Here is what each one uses, what leaves our systems, and what's kept.

Transcribe

What it does
Writes down what's said in a video or recording, with times, so you can read along, jump to a moment and search by what was said.
Model
Whisper (large-v3-turbo), an open speech-recognition model made by OpenAI, run by Cloudflare Workers AI. OpenAI itself never receives anything.
What leaves our systems
The soundtrack only, cut into short audio pieces and sent to Cloudflare Workers AI. No pictures, names or other details.
What the provider keeps
Cloudflare doesn't store the audio or the text it returns, and doesn't train on them.
What we store
The transcript, in our database with the file. Anyone who can see the file can read it, and it's deleted with the file.

Find faces

What it does
Finds faces in a video and groups the same person's faces together, so you can name someone once and find every clip they're in.
Model
YuNet (finds faces) and SFace (describes them), two open models published by OpenCV. They run on our own background jobs, not at any AI company.
What leaves our systems
Nothing. Video frames are read on our own systems and never sent to an outside AI service.
What the provider keeps
No outside AI company is involved.
What we store
A small image of each face found, in our storage, and a list of 128 numbers describing that face (an “embedding”), in our database. Embeddings are compared to group faces of the same person. They are biometric data, and they never leave our systems. People's names and birthdates, added by you, are stored with them. Faces are deleted with their file (apart from the off-site backup, under “Deleting things”), and you can delete or move any face on a person's page.

Dates and tags

What it does
Works out when a clip was filmed (from date stamps, what's said, birthdays and people's ages) and suggests tags. Suggested tags wait for you to accept or reject them, and a date you set by hand is never changed.
Model
Claude Haiku 4.5, made by Anthropic, through Anthropic's API.
What leaves our systems
For each video: six still frames spread through it, strips from the top and bottom of other frames (where date stamps sit), the transcript if there is one, the first names of people already named in the clip, and the library's tag names. The frames show the people in the video, children included. When someone in the clip has a birthdate, a few frames carry numbered boxes around faces so Claude can judge ages. Claude is never told whose face is in a box, and birthdates are never sent: Claude answers with ages, and we work out dates ourselves.
What the provider keeps
Anthropic doesn't train on data sent through its API and deletes it within 30 days.
What we store
Date clues (with a sentence on what Claude saw or heard), suggested tags, and a copy of any boxed frames so a clue can be checked. All in our own storage and database, deleted with the file. The other frames aren't kept.

Deleting things

Our database provider keeps short-lived backups for recovery, so deleted data can take a few days to leave them.

So that nothing is lost to a mistake or an outage, hawmie also keeps an off-site backup with AWS. Each night it copies every new file and face image, including those in the trash, and a copy of the database (titles, tags, dates, transcripts, people and face data). Deleting a file in hawmie, or emptying it from the trash, doesn't remove its backup copy: files stay in the backup, and database copies are kept for a year. A file deleted before that night's backup is never copied. To have something removed from the backup too, email jonchaney@gmail.com.

Changes

If this policy changes in a way that matters, you'll be asked to read and agree to it again the next time you sign in. Questions: email jonchaney@gmail.com.

Privacy · Terms