Overview

Alcoves is a self-hosted, privacy-first collaborative media library — a Google Drive plus Google Photos replacement you run yourself. It stores your photos, videos, audio, and documents; it understands them (faces, objects, sounds, speech); it lets a small group browse, organize, and share them — and it does all of this on hardware you own, with no telemetry, no per-seat pricing, and no third party in the loop.

These are starter docs

This documentation site is new. The pages here are concise samples derived from the project’s in-repo docs and vision; they’ll be expanded into the full reference over time.

Why Alcoves

The convenient tools for managing personal and family media are the ones that mine it. The tools that respect your privacy are usually too primitive to be worth using. Alcoves refuses that trade-off: it aims to be as capable as the cloud incumbents on the features that matter, while being something you fully own.

The pillars

Own your media

Local-disk or S3-compatible storage, your database, your queue. The Go binary is a pure API — nothing phones home. A Raspberry Pi with a USB drive and a rack with an object store are both first-class.

Understand it locally

An async pipeline does SHA-256 dedup, thumbnails, video proxies, waveforms, and a full suite of CPU-only ONNX ML — faces, objects, audio events, and speech transcription. Models download on demand. No GPU. No inference leaves the box.

Collaborate with real roles

Libraries are the unit of sharing. Owner / admin / viewer roles, invite links with caps and expiry, and a per-library access gate enforced in middleware before any handler runs.

Video is first-class

A real timeline editor: zoomable waveform, draggable named “moments”, transcript search, audio-detection overlays, MP4 export, and public, OG-embeddable share links.

Who it’s for

• Individuals who want a private home for a lifetime of photos and video, with real search and real organization.
• Families sharing a library — grandparents, kids, events — with sensible roles so not everyone is an admin.
• Small teams and creators (1–100 users per instance) collaborating on video: clipping, transcribing, tagging, and sharing highlights.
• Self-hosters and homelab operators who want a clean Helm chart and Docker images for the Postgres, the queue, and the GPU-less inference.

Alcoves is explicitly not built for the 100,000-user multi-tenant SaaS case. It optimizes for a trusted instance shared by a known, bounded group.

Next steps

Quickstart Run the full stack locally with Docker Compose in a few minutes.

Configuration The ALCOVES_* environment variables that drive every instance.

Architecture How the Nuxt frontend and Go/Echo/GORM/Asynq backend fit together.

Privacy & local AI Why every model is CPU-only and how inference stays on your box.