Privacy

What this site and this software do with your data. Short version: almost nothing.

A project whose argument is that your companion should live on your own machines would be a poor one if the website quietly did the opposite. So this page is specific rather than reassuring, and it names the places where data does move — including the ones outside our control.

This website

The web server keeps a deliberately blunt request log, and never records your IP address. What is written down is the path you asked for, the response status, the size, the user agent, and a truncated network address: the final octets are zeroed before anything is stored, so 203.0.113.47 is only ever recorded as 203.0.113.0.

That is enough to tell us which country or region a download came from, and whether a 1.8 GB installer completed or a link is quietly returning 404. It is not enough to pick you out, follow you between visits, or build a profile — and there is nothing here that could be joined against another dataset to re-identify you, because we hold no other dataset. Logs roll at 10 MB and only the last five files are kept.

No cookies. No analytics. No third-party requests of any kind — the pages load their own stylesheet and nothing else, so nothing about your visit reaches another company.

One exception, and we would rather name it than let you find it. Blocking an abusive source requires knowing which source it is, and an address with its last octets removed cannot be blocked. So a second, separate log does record the full address — and is deleted within about a day. Nothing reads it except the tool that decides whether to ban an address; it is never aggregated, never joined to the log above, and never backed up. If we stop running that tool, the log goes away rather than sitting there unread.

Three honest limits. Truncating an address reduces what we hold, and is the standard way of separating "roughly where" from "who" — but privacy law in some places still treats even a shortened address as personal data, so we describe it rather than claim an exemption. The short-lived abuse log above is a real, if brief, exception to "we never record your IP". And the machine runs at a hosting provider, which can see network traffic at its own layer regardless of what we log. That last one is true of every site on the internet; we mention it because statements about logging are otherwise easy to over-read.

The software

Wyrdsekai has no telemetry, no analytics, no crash reporting, and no phone-home. It does not check in, count installs, or report usage. Your companion's memories, your conversations, and your world live on disk on your own machines.

Nothing in the default configuration sends your data anywhere. Three things can change that, all of them your choice, and all of them worth understanding:

• Cloud inference. If you configure an API backend instead of running a model locally, your conversations go to that provider under their terms and privacy policy. We never see them, and we cannot make promises on that provider's behalf. Running locally is the default for exactly this reason.
• Someone else's relay. A relay is how a phone reaches a household without opening a port. Traffic through it is encrypted end to end, so the operator cannot read what passes — but they can see connection metadata: that a device connected, when, and roughly how much. Run your own if that matters to you; the relay bundle is a download on this site.
• Skills and tools you enable. If you connect your companion to a calendar, a mail account, a search service or an external tool, it talks to those services. That is the point of connecting them, and each is governed by its own provider.

Model downloads

Model weights are fetched from this site and, for some models, from Hugging Face. Requests here are logged as described above — path and truncated address, never your IP. Requests to Hugging Face are ordinary web requests to them and fall under their privacy policy.

Email

If you write to wyrd@wyrdsekai.org we keep the correspondence so we can answer it and remember the context. We do not add you to anything, because there is nothing to be added to.

The public relay we run

We operate a relay at relay.wyrdsekai.org, so that people without a spare box with a public address can still reach their household from a phone. Using it is optional and creates no account; the relay page covers how it works and how to leave.

We cannot read what passes through it. A relay moves encrypted bytes between your phone and your household; envelopes are signed end to end and the tunnel terminates at your devices, not at us. That is a property of the protocol rather than a promise about our behaviour, which is the only sort of assurance that is worth anything.

What we do see is metadata, and you should assume we see all of it: which identity connected, when, from which network address, roughly how much traffic it moved, and which subjects were routed. No relay operator can avoid this. It is kept as a registration record — identity key, tier, liveness timestamps — for as long as the registration is active, and pruned when it lapses: 24 hours at the newcomer tier, seven days above it. None of it is analysed, profiled, or shared, and there is nobody to share it with.

If that pattern-of-life is itself sensitive to you, run your own relay. The bundle is a 115 KB download and takes three commands. We would rather say this plainly than have you discover the trade-off later.

Mobile apps

The Android and iOS clients are not currently distributed through app stores; you build and sideload them. If that changes, the stores themselves collect their own data about downloads under their own policies, and this page will be updated to say so.

Changes

If any of the above stops being true — logging turned on, an app published, the relay's terms changed — this page changes first, and the change is described rather than silently applied. The software is open source, so you are never dependent on this page being honest: you can read what it actually does.