MinIO vs Garage
MinIO vs Garage compared on project status, licensing, hardware, durability, topology, S3 API coverage, access control, replication and pricing, using MinIO's archived community repo, AIStor docs and pricing, and Garage v2.5.0 as of October 2026.
Quick Answer
Pick Garage for self-hosted S3 on modest or mixed hardware: one dependency-free binary, 1 GB RAM per node, three-way replication across zones that can sit in different homes or data centres over the Internet, static website hosting and an admin API. Accept what it lacks: no object versioning, no object lock, no bucket policies or ACLs, and no erasure coding. Do not start a new MinIO community deployment: the GitHub repo is archived and marked no longer maintained, and the community edition was source-only before that. MinIO's maintained product is AIStor: AIStor Free is a full-featured single-node edition, and multi-node Enterprise Lite and Enterprise are quote-only, for teams that need erasure coding, versioning, object lock and replication at scale.
Last updated October 28, 2026. Covers the archived MinIO community repo (final release RELEASE.2025-10-15T17-29-55Z), AIStor docs and pricing tiers, and Garage v2.5.0 (24 Oct 2026) with its feature, S3 compatibility and deployment docs, as of 28 Oct 2026.
| Feature | MinIO | Garage |
|---|---|---|
| Licence | Community: AGPL-3.0 (archived); AIStor: MinIO Software License | AGPL-3.0 |
| Project Status (Oct 2026) | Community repo archived, no longer maintained; AIStor actively developed | Active; publicly funded (NGI, NLnet) |
| Latest Release | Community: RELEASE.2025-10-15T17-29-55Z (final); AIStor: commercial releases | v2.5.0 (Docker image 24 Oct 2026) |
| How You Get It | Community: source only; AIStor: download with licence key | Single static binary; Docker image dxflrs/garage |
| Hardware Guidance | AIStor production: 8+ drives and 128 GB+ RAM per host | 1 GB RAM, 16 GB disk; x86_64, ARMv7, ARMv8 |
| Data Protection | Reed-Solomon erasure coding (EC:3+ for production) | Full-copy replication (replication_factor 3 recommended) |
| Cluster Topology | Server pools of erasure sets; AIStor Free is single-node only | Zones across sites over the Internet; mixed node sizes |
| S3 API Coverage | Broad (MinIO publishes no public endpoint list) | Core, multipart, CORS, website, presigned URLs, SSE-C |
| Versioning / Object Lock | Yes: versioning, object lock, retention | No |
| Access Control | IAM-style policies, service accounts, STS, LDAP | Per-key, per-bucket permissions; no policies or ACLs |
| Static Website Hosting | Not documented on pages read | Yes |
| Server-Side Encryption | Yes, with key management server (AIStor) | SSE-C only |
| Replication | Active-active, bucket, batch and site replication (AIStor) | Built-in cross-zone copies; no S3 bucket replication API |
| Pricing | AIStor Free $0 (single node); Enterprise Lite and Enterprise: quote 5/10 | Free (AGPL-3.0) 10/10 |
| Support | Community: best effort; AIStor Enterprise: 24/7, <4 h SLA | Community project |
MinIO
High-performance, S3-compatible object storage from MinIO, Inc. The AGPLv3 community edition on GitHub is archived and marked no longer maintained; the company now ships AIStor under the MinIO Software License, with a free single-node edition and quote-based multi-node Enterprise Lite and Enterprise subscriptions.
Pros
- Erasure coding with configurable parity: erasure sets of 2 to 16 drives by default (up to 32 on recent AIStor releases), with EC:3 or higher recommended for production
- Broad S3 feature set in AIStor: versioning, object locking and retention, encryption with a key management server, and active-active, bucket, batch and site replication
- Built for large deployments: server pools, horizontal scaling and accelerators listed for AVX-512, AVX-2, NEON, SVE and VSX
- AIStor Free gives a full-featured single-node deployment at no charge, with community Slack and docs support
- Commercial support with SUBNET on Enterprise: 24/7 direct-to-engineer support with a sub-4-hour SLA, plus FIPS 140-3 and SOC 2 listed
Cons
- The community repo is archived and its README says it is no longer maintained; the last community release is RELEASE.2025-10-15T17-29-55Z
- The community edition went source-only before that: no new prebuilt binaries or images, so you build with go install or the Dockerfile
- AIStor is licensed under the MinIO Software License, not AGPL, and the docs say production clusters need an active licence
- Multi-node AIStor (Enterprise Lite under 400 TiB, and Enterprise) has no public price; you request a quote or start a 60-day trial
- Production hardware guidance is heavy for a homelab: 8 or more drives and 128 GB or more of memory per host
Best For
Teams storing tens of terabytes and up who need erasure coding, versioning, object lock, replication and vendor support, and who are fine with a commercial licence; or a single node on AIStor Free.
Garage
An AGPLv3, S3-compatible object store written in Rust by the non-profit Deuxfleurs, built for small self-hosted clusters spread across several sites on ordinary Internet links. It ships as a single dependency-free binary, replicates whole copies of data across zones, and has been funded by NGI POINTER and NLnet grants.
Pros
- Low requirements: any x86_64 CPU from the last decade or ARMv7/ARMv8, 1 GB RAM and 16 GB of disk per node
- Geo-distribution is the point: three copies placed in different zones, working over links up to 200 ms latency and down to 50 Mbps
- No Raft or other consensus on the request path, so requests do not wait on a leader node
- Heterogeneous hardware is fine; the cluster layout uses each node's zone and capacity and rebalances when nodes are added or removed
- Extras that MinIO does not document: buckets served as static websites, bucket aliases (rename without copying), deduplication and optional Zstd compression
- Admin REST API, Prometheus metrics, OpenTelemetry traces, and Kubernetes and Nomad node discovery
Cons
- No object versioning and no object lock or retention, so no S3-level immutability for backups
- No bucket policies or ACLs; access is Garage's own per-key, per-bucket permission system
- Replication only, no erasure coding by design, so three-way replication stores three full copies
- Server-side encryption is SSE-C only; the project suggests disk or client-side encryption instead
- Lifecycle rules support only expiration and aborting incomplete multipart uploads, and unsupported endpoints return 501 Not Implemented
Best For
Homelabs, small hosting collectives and small teams that want durable S3 for backups, media and static sites across two or three locations on cheap or second-hand hardware.
What happened to MinIO's community edition
The minio/minio repository on GitHub is now archived. Its README opens with 'THIS REPOSITORY IS NO LONGER MAINTAINED' and points to two alternatives: AIStor Free, a full-featured standalone edition for community use, and AIStor Enterprise, the distributed edition with commercial support. Before the archive, the README had already moved the community edition to source-only distribution: no new prebuilt binaries, with go install or a self-built Docker image as the install paths and old binaries left online without updates. The last community release on GitHub is RELEASE.2025-10-15T17-29-55Z, which fixed a privilege-escalation advisory in service accounts and STS session policies. Commits after that only changed the README, the last in February 2026. The docs site now redirects to AIStor documentation, which says AIStor is licensed under the MinIO Software License and production clusters require an active licence. The pricing page lists three tiers: Free (single-node, community support), Enterprise Lite (multi-node, for capacity under 400 TiB, Lite SUBNET, optional support add-on) and Enterprise (multi-node, 24/7 support with a sub-4-hour SLA). Only Free has a price; the others are 'request pricing' with a 60-day trial. For self-hosters this means the AGPL MinIO you may have deployed years ago will not get new fixes, and the maintained MinIO is a proprietary product.
Durability: erasure coding versus whole-copy replication
MinIO splits each object into data and parity shards across an erasure set. AIStor docs describe erasure sets of 2 to 16 drives by default (up to 32 with MINIO_ERASURE_SET_DRIVE_COUNT on releases from February 2026), parity from EC:0 up to half the set, and a firm production rule: EC:3 or higher. A 16-drive set at EC:4 keeps 12 data shards per object, so it survives four lost drives while storing roughly a third more than the raw data, far less overhead than three full copies. The trade-off is hardware: uniform nodes, dedicated local drives and, in AIStor's production guidance, 8 or more drives and 128 GB or more of memory per host. Garage rejects erasure coding on purpose. Its design-goals page lists storage optimisations as a non-goal because coding makes data placement and synchronisation harder; it limits itself to duplication. With replication_factor = 3, every block is stored on three nodes in three different zones, and the real-world deployment guide warns that usable capacity is bounded by those placement rules. Garage also deduplicates blocks and can compress them with Zstd, which recovers some space for repetitive data. In practice: for a few terabytes across three cheap boxes, three copies cost little and are easy to reason about. At hundreds of terabytes on dense servers, erasure coding saves real money in drives.
Topology: one fast cluster versus nodes in different places
Garage was built by Deuxfleurs, a non-profit host without its own backbone, so it assumes nodes sit in different homes, offices or data centres connected over the public Internet. Its homepage targets links of 200 ms or less and 50 Mbps or more, and it uses no Raft or other consensus on the request path, which keeps latency between sites from stalling every write behind a leader. You assign each node a zone and a capacity in the cluster layout; Garage places copies across zones and rebalances when the layout changes. Mixed and second-hand hardware is explicitly supported. Its own production cluster is 9 nodes in 3 locations, and Triplebit runs 15 storage nodes in 3 locations, per the use-cases page. MinIO is designed around server pools of uniform hardware with erasure sets spanning drives on those nodes, which suits a rack or a single data centre. AIStor adds multi-site replication (active-active, bucket, batch and site) to link separate deployments, rather than stretching one erasure set across slow links. A single node is possible with both: Garage's quick start runs garage server --single-node with replication_factor = 1, and says not to use that in production. AIStor Free is single-node by definition, so any multi-node MinIO you build today is either the unmaintained community code or a paid AIStor subscription.
S3 compatibility: what Garage leaves out
Garage publishes a detailed compatibility table, and it is worth reading before you migrate. Implemented: signature v4, path-style and vhost-style URLs, presigned URLs, SSE-C, all core object and bucket calls, ListObjectsV2, PostObject, the full multipart set including UploadPartCopy, CORS and the website endpoints. Missing: bucket versioning (GetBucketVersioning is a stub that always reports disabled), ListObjectVersions, object lock, legal hold and retention, bucket and object ACLs (GetBucketAcl is partial), bucket policies, replication endpoints, SSE-S3 or KMS encryption, notifications and tagging. Lifecycle rules work only for Expiration and AbortIncompleteMultipartUpload. Anything missing returns 501 Not Implemented. Most self-hosted software is fine with that subset: restic, rclone and duplicity backups, Nextcloud, PeerTube, Mastodon and Matrix media, Nix caches and static site hosting all appear in Garage's integration docs or use cases. Where it bites: backup tools that rely on object lock for ransomware-proof immutability, apps that expect bucket policies for public-read access (Garage uses website hosting or its own permission system instead), and anything that writes object tags. MinIO does not publish an endpoint-level compatibility list, a gap Garage's page also notes, but AIStor's documentation and feature list cover versioning, object lock with retention, encryption with a key server and replication. If those features are requirements, that settles it.
Running costs and migration
Garage costs nothing in licences, and it is easy to start: one binary or the dxflrs/garage Docker image, a TOML config with an rpc_secret shared by every node, the default LMDB metadata engine (or SQLite, which the guide suggests if you cannot rely on filesystem snapshots for LMDB, and on 32-bit platforms), then garage layout assign with a zone and capacity per node. The admin API covers nodes, keys, buckets and aliases, so it can be automated. The latest image is v2.5.0, published 24 Oct 2026, following v2.4.0 and v2.4.1 in September. MinIO's cost depends on which MinIO you mean. Community MinIO is free and AGPL, but frozen. AIStor Free is $0 for one node under the MinIO Software License, so read those terms before using it commercially. Multi-node AIStor is a quote, with Enterprise Lite aimed at under 400 TiB. Budget the hardware AIStor recommends too. Moving between them is an S3-to-S3 copy: create buckets and keys on the target, then use rclone or a similar tool to sync, and repoint applications by changing the endpoint and credentials. Check versioned buckets first: Garage stores only the current version of each object, so older versions and lock settings on a MinIO bucket will not come across.
Verdict
For a new self-hosted S3 store on ordinary hardware, choose Garage. It runs on 1 GB of RAM, keeps three copies of each block in different zones, tolerates slow links between sites, and covers what most self-hosted apps actually call: core object operations, multipart uploads, presigned URLs, CORS and static websites. Check one thing first: if your backup tool or compliance rules need object versioning or object lock, Garage cannot provide it today. Do not build anything new on MinIO's community edition. Its GitHub repository is archived and labelled no longer maintained, and there are no new community binaries, so you would be running unpatched code from October 2025 onwards. If you already run it, plan a move. Choose MinIO AIStor when you need erasure coding, versioning, object lock and replication on large, uniform hardware, and you have budget for a commercial licence and support: AIStor Free covers one node, Enterprise Lite and Enterprise are quote-only. If you need versioning or object lock across several nodes without a commercial licence, neither fits; look at other open-source S3 servers instead.