FAQ

Do I need a GPU?

No. All media analysis in Alcoves — face and object detection, audio tagging, speech transcription — runs on CPU by design. See Privacy & local AI.

Is any of my data sent anywhere?

Your files and everything derived from them stay on your instance, and there is no analytics or usage tracking. The few outward connections are ones you control: ML models download on first use (mirrorable), the map view loads OpenStreetMap tiles by default (configurable), and error reporting only exists if you configure a Sentry DSN. The details are on the privacy page.

What hardware do I need?

An x86_64 machine — the published Docker images are x86_64-only right now. 8 GB of RAM is comfortable. Transcription is the biggest memory consumer: the default whisper model wants several GB, and the admin panel lets you choose a smaller one for lighter hardware. Storage is a local disk (or any volume you mount); size it for your library.

How do I deploy it?

With Docker Compose on a single machine, or the Helm chart on Kubernetes. You bring PostgreSQL (with pgvector) and a Redis-compatible queue (Dragonfly works well); one published image runs the frontend, the API, and the worker. Start with the Quickstart.

Who can administer an instance?

The first registered user becomes the instance owner. Only the owner can open the admin panel, the job-queue dashboard, registration policy, and ML model selection. Within libraries, the roles are owner / admin / viewer.

Is it production-ready?

Alcoves is alpha (0.x). It works and is used daily, but expect breaking changes between releases — they ship with migrations and changelog entries. It is built for bounded, trusted groups, not anonymous mass signup.

Can other tools talk to it?

Yes — Alcoves includes an MCP (Model Context Protocol) server, so MCP-capable assistants and scripts can browse libraries, upload, and organize files. It runs over stdio or HTTP, authenticated with a personal access token or OAuth. See the MCP server page.