Release notes from the survey office.

Continuous early access builds

Unofficial early-access desktop builds for Windows, macOS and Linux. The artifacts include the Electron desktop runtime and app dependencies. Local AI is optional and built in: the app can download a small model onto its bundled llama.cpp runtime, or connect to an Ollama, LM Studio or llama.cpp server you already run. The generator itself never needs a model.

New here? Start with the first region quickstart, then come back for downloads and known issues.

View SHA-256 checksum files, updater metadata, and all mirrored release assets.

These builds are unsigned, so every OS challenges the first launch. First-run steps for each platform and the current known issues are below.

What will bite you first

Continuous builds are working software with rough edges. This is the honest list. Anything you hit that is not here belongs on the issue tracker — seed and app version included, please, and say it is Hexes.

The builds are unsigned — every OS challenges the first launch

Windows: SmartScreen shows “Windows protected your PC.” Click More info, then Run anyway. This appears once per download.

macOS: the first open is blocked because the app is not notarized. Open System Settings → Privacy & Security, scroll to the message about the blocked app, and click Open Anyway. If macOS instead claims the app is “damaged,” clear the quarantine flag in Terminal: xattr -dr com.apple.quarantine "/Applications/Hexes on Automatic.app". In-app updates also do not install on macOS while builds are unsigned — when the app offers an update, re-download from this page instead.

Linux: mark the AppImage executable (chmod +x or the file manager's Properties → Permissions), then run it. Some distributions need the FUSE 2 library (libfuse2) for AppImages.

Current known issues

• A region is reproducible from its seed within an app version. Across versions the generator changes, so keep the Save JSON export (or the vault's state/ file) for any region you want back exactly.
• Export to Vault and Open from Vault need the desktop app; a plain browser build can only generate, polish and save JSON. That is a boundary, not a bug, but it surprises people.
• The built-in local-AI setup (bundled llama.cpp runtime with a one-click model download) is new, and its first-run flow on a completely fresh machine is still being verified. If it misbehaves, the Ollama / LM Studio connection path is the fallback — and a bug report is a favor.
• Director rumors can occasionally mention a place by its pre-rename name when both were written in the same polish pass. Regenerating the polish usually clears it.

Application changelog

Every pull request merged into the application, newest first. Entries arrive automatically with each build; the release page carries the full notes.