Snap name: harbour-ai
Summary: HARBOUR AI is a local multi-agent AI desktop app for UK businesses. It runs entirely on-device via Ollama — no cloud, no telemetry. It includes a FastAPI backend, Next.js frontend, and Electron shell, and ships as a single desktop application.
Why classic confinement is needed:
HARBOUR AI requires access to several system resources that cannot be satisfied within strict confinement:
- Ollama integration — the app communicates with Ollama, which runs as a system-level service. Strict confinement does not reliably allow the inter-process communication required.
- Filesystem access — the app reads and writes a SQLite database, a knowledge base, user-uploaded documents, and configuration files across the user’s home directory. The home interface does not cover all required paths.
- xdotool / active window detection — the Screen Memory feature reads the active window title using xdotool. This requires access to the X display that is not available under strict confinement.
- Audio input — the voice commander feature uses the system microphone. While the audio-record interface exists, it does not function correctly within the Electron context under strict confinement.
- Electron shell — Electron apps have well-documented compatibility issues with strict snap confinement due to their Chromium base and IPC mechanisms.
HARBOUR AI is a desktop productivity application with no need to access other users’ data or system internals beyond the above. Classic confinement is the standard approach for Electron desktop apps of this type (VS Code, Slack, Discord all use classic confinement).
The snap name harbour-ai is registered and CI is configured to auto-publish on every release once confinement is approved.
Happy to provide any additional information.