Garage

Key features

• S3-compatible object API covering the common surface: signature v4, path-style and virtual-hosted-style URLs, presigned URLs, and SSE-C server-side encryption with customer keys.[1][2]

• Geo-distributed replication across distinct physical locations: copies of data live in several places at once, typically three, so the cluster survives network, power, or hardware failures at one site.[1][2]

• Flexible topology with automatic rebalancing: storage nodes join or leave over time and Garage redistributes data itself to keep the desired number of copies.[1]

• Configurable replication modes and consistency levels: replica count and consistency adapt to the workload, with replication_factor = 3 the documented usual choice.[1][2]

• Block-level deduplication with optional Zstd compression: stored blocks are deduplicated, and compression can be switched on to trade CPU for disk.[1]

• Standalone self-contained deployment: one dependency-free binary with no external service required, installable from release binaries, distribution packages, Docker, systemd units, Kubernetes manifests, or a source build.[1][2]

• Static website hosting served directly from buckets: Hugo deploys with integrated publishing logic, Publii speaks Garage natively as a custom S3 provider, and other generators sync through any S3 CLI.[1]

• Cluster operations through a CLI and dedicated admin API: layout changes, bucket and key management, durability repairs, and monitoring run through purpose-built operator tooling rather than the S3 endpoint.[1][2]

Who will love it

• Teams spreading storage across homes, offices, and datacenters: geo-aware placement keeps copies in distinct locations and tolerates consumer-grade links and power at any single site.[1][2]

• Cost-conscious builders serving backups and static sites from owned hardware: zero license cost, restic and duplicity fit the backup slot, and buckets double as website origins.[1][2]

• Developers already speaking S3 who need a small-values sidecar: the compatible subset covers standard object workflows while K2V handles many small key-value entries the S3 shape serves poorly.[1][2]

Who should avoid it

• Teams depending on the full S3 surface: bucket versioning and signature v2 are missing, and unimplemented endpoints answer 501, so clients with wide call surfaces need verification before committing.[1]

• Single-server minimal-ops shops: learning-only single-node setups carry no redundancy, real deployments start at three separately located nodes, and the catalogue rates deployment moderate.[1][2]

• Non-technical users wanting storage without servers: configuration files, key management, and cluster operations assume operator skill, and there is no bundled graphical console or managed offering in the inspected docs.[1]

Alternatives and compatibility