Desktop_Packaging

Desktop Packaging

Orrin ships as a native desktop app: a PyInstaller-frozen runtime in a pywebview window, with no browser, no Node at runtime, and no open port. packaging/ holds the build machinery. Current builds are unsigned — expect OS trust prompts.

The native shell

• Window: pywebview loads the built UI (frontend/dist) from disk and talks to the brain over the in-process js_api bridge (backend/server/bridge.py) — no HTTP port at all in bridge mode.
• Fallback: if pywebview is unavailable, main.py falls back to a loopback API plus a browser tab (one OS-assigned port).
• Tray: backend/server/tray.py provides the system-tray presence.

Per-user data

A packaged app must not write inside its own bundle. The frozen app sets ORRIN_DATA_HOME so all persisted state (both the brain/data/ and root data/ trees) relocates to a per-user data directory. The same variable works for source runs.

Secrets

A shipped app never carries secrets in its bundle or a plaintext .env. Keys the user pastes into Settings live in the OS keychain — Keychain (macOS) / Credential Manager (Windows) / libsecret (Linux) — via keyring. brain/utils/secrets.py is the one module that reads/writes them; the backend's POST /api/settings and main.py's boot both go through it.

Build machinery (packaging/)

• orrin.spec — the PyInstaller spec for the frozen build.
• bundle_models.py — pre-bundles the embedding models so the app runs offline (no first-run download).
• entitlements.plist — macOS entitlements for the (future) signed build.
• linux/ — Linux packaging variants.
• packaging/README.md — build instructions and logs from CI watches.

Product surfaces

The packaged app carries the full product: the UI rooms (see Face & Brain UI), Settings with keychain-backed keys, the existence model with its end-of-life screen (see Existence and Lifecycle), and self-code support — all with the same symbolic-first, fail-closed behavior as a source run.

Known limitations

• Builds are unsigned (macOS Gatekeeper/SmartScreen warnings).
• The full dependency set is heavy (PyTorch + spaCy); there is no low-resource install profile yet.

Code pointers

• packaging/ — spec, model bundling, entitlements
• main.py — bridge-vs-fallback UI selection at boot
• brain/utils/secrets.py — keychain storage