Which Community Edition projects are gaining ground. One ranking signal, one window: 180-day pull growth, with upvotes, comments, grade upgrades, and HN / Reddit mindshare layered on top. Pull counts aggregate every variant a publisher ships, so architectures and tag families roll up to one project number. The 30-day column is just a shorter-window read; it does not drive rank. See /about/grades.
End-user self-hostable applications only. Language runtimes and base layers live in building blocks. · Updated 2026-11-14 07:55 UTC
NATS (Apache-2.0, CNCF) is the pick for microservice-to-microservice messaging where you want persistence (JetStream), key/value, and object storage in one binary without running Kafka. Single-digit MB container, millions of messages per second on commodity hardware. Pick RabbitMQ when you need rich routing semantics; pick Kafka when you need log replay at TB scale.
Solr (Apache-2.0) is the search engine the Lucene core team runs; shares the engine with Elasticsearch but a different API and distribution model. Enterprise sites and library catalogs still run Solr for its mature faceting, highlighting, and MLT (more-like-this) recommendation. For log analytics or real-time ingest, Elasticsearch/OpenSearch now dominate. Picks over them when the workload is "classic enterprise search" and you already have ops familiarity.
Ombi is the self-service layer in front of the arr stack: users log in with their Plex account, search for a title, click request. Radarr/Sonarr get the job via Ombi's integration. Admins can require approval or auto-approve by media type. Overseerr (Jellyfin-focused) is the newer alternative with a prettier UI.
etcd (Apache-2.0, CNCF graduated) stores small, critical data with strong consistency via Raft. Kubernetes stores its entire API state in etcd; so do Rook, Patroni, and many Vault setups. Standalone, it's a solid service-discovery and config store, especially for microservices that need linearizable reads. Not for high-throughput writes or large payloads — small values, lots of watchers is the sweet spot. Alternatives: Consul (HashiCorp, more features), ZooKeeper (older, JVM).
MinIO (AGPL-3.0) is what teams reach for when they want object storage without the AWS bill or latency: on-prem, on your own disks, with the same S3 API your SDK already uses. Deploy a single-node for backups and dev; cluster across hosts for production with erasure coding. Pairs well with Mastodon media storage, Loki log blobs, and Postgres WAL archives. Alternatives: SeaweedFS (lighter), Ceph (richer, heavier), native S3 (when the cloud bill fits).
Loki stores raw logs cheap (object storage) and indexes only the labels you declare. Teams use it when ELK's cost or operational weight is too much and when they already run Grafana for metrics. The LogQL query language borrows directly from PromQL, so SREs who speak Prometheus pick it up in minutes. Reach for Elasticsearch instead when you need full-text search over log bodies.
Postgres is the "safe choice" default for teams starting a new backend. Operators pick it over MySQL for its JSONB performance, better data integrity defaults, and extensions ecosystem (pgvector for embeddings, PostGIS for geo, Timescale for time-series). Reach past it only when you have a specific reason: a hard write-throughput ceiling, a multi-region active-active requirement, or an existing MySQL team.
MySQL remains the pragmatic choice when you're wiring up something that assumes it: WordPress, Magento, phpMyAdmin, old Rails apps, SaaS products with "connect to your MySQL" as the first-class path. New projects increasingly default to Postgres; MySQL still shines when ecosystem compatibility (Vitess, PlanetScale, cloud managed services) outweighs the engine-level trade-offs.
WordPress is the default when the brief is "I need a website I can edit without touching code." Hosted at WordPress.com for the no-ops path, self-hosted via this image for everything else. Reach for a static site generator (Astro, Hugo) when the content barely changes; keep WordPress when non-developers maintain the content. Plugins and themes are both strength and attack surface — keep them minimal and up to date.
Traefik (MIT) listens to your container orchestrator (Docker, Compose, Kubernetes, Nomad, Consul) and configures itself from service labels. The reason it caught on: new service with the right labels appears, Traefik immediately routes to it with a Let''s Encrypt cert. For hand-tuned configs on a traditional VM, Nginx or HAProxy still win on footprint; for container-native stacks, Traefik is the pragmatic default.
Ghost is the "I want to run my newsletter myself" default — the MIT-licensed foundation also powers Ghost's hosted version. Writers pick it over Substack when owning the subscriber list and avoiding the 10% platform cut matters. Technically it's a Node.js app with a MySQL/SQLite backend; this image handles both. Alternatives: Buttondown, Beehiiv (SaaS), WriteFreely (minimal, Fediverse-native).
Distribution (the project behind `registry:2`) is the reference OCI registry. Teams run it to host company-internal images without a cloud bill, to mirror the public registries behind a proxy for faster builds, or as a cache layer in bandwidth-constrained environments. For richer features (vulnerability scanning, retention policies, replication, UI) look at Harbor, which bundles Distribution with its own control plane.
Valkey (BSD-3-Clause) is where teams land after Redis moved to the SSPL and Elastic License in 2024. Deployed exactly like Redis: same client libraries, same cluster semantics, mostly identical config. Reach for it over Redis for every new project; migrate existing Redis workloads at your next upgrade window. KeyDB and DragonflyDB are the "same-protocol, faster/different-architecture" alternatives.
cloudflared (Apache-2.0) establishes an outbound tunnel to Cloudflare and routes incoming requests from your Cloudflare-managed hostname through it. Nothing on your network listens on a public port; no firewall holes. Self-hosters pick it when ISP restrictions or NAT make port-forward impossible; teams pick it to add Cloudflare Access controls in front of internal apps. Free tier covers most personal use; business use may push you to Cloudflare for Teams.
Gitea (MIT, forked from Gogs) is a single Go binary that gives you git over SSH + HTTP, pull requests, issues, CI integration, package registries, and a GitHub-shaped UI. Teams run it when GitLab is too heavy for the hardware or when the only requirement is "a git host that isn't github.com." Forgejo (hard fork, same code lineage) is the drop-in alternative when you prefer Codeberg-adjacent governance. SourceHut is the austere alternative.
Prometheus (Apache-2.0, CNCF) is the metrics half of every modern SRE stack. Services expose a /metrics endpoint; Prometheus pulls on a schedule, stores locally, and alerts on PromQL queries. For long-term retention you add Thanos, Cortex, or VictoriaMetrics; for logs you add Loki or Elasticsearch. The "pull" model fights "push" model systems (statsd, Datadog agent) — pick the one your infrastructure already assumes.
Radarr is one node in the "arr stack" — Sonarr for TV, Lidarr for music, Bazarr for subtitles, Jackett for indexer glue, SABnzbd/NZBGet/Transmission as the download clients. Self-hosters set it alongside Plex, Jellyfin, or Emby to curate a personal library automatically. LinuxServer.io's image is the canonical community build; the official project ships Radarr itself but relies on LinuxServer for container distribution.
Caddy (Apache-2.0) is the "modern defaults" web server — HTTPS by default, HTTP/3 by default, a config language (Caddyfile) that reads like English. Reach for it over Nginx when a single developer runs the ops and spending an afternoon fighting cert-manager is the wrong trade-off. Nginx still wins on raw QPS and ecosystem depth; Caddy wins on time-to-first-TLS and config readability.
Nexus Repository OSS (Eclipse Public License 1.0 — the Sonatype-maintained OSS core) stores Maven, npm, PyPI, NuGet, Helm, Docker, and raw-binary artifacts. Teams run it as the single artifact store that CI publishes to and build-time dependencies resolve from. Proxy caching of upstream registries (Maven Central, npm, PyPI) also cuts build times. Pro version adds HA, staging, firewall rules. Alternatives: JFrog Artifactory (JCR free, Pro proprietary), Harbor (OCI-only but strong on container governance).
Kafka (Apache-2.0) is the pick when you need durable, replayable, high-throughput event streams: audit logs, change-data-capture pipelines, event sourcing. Writes are ordered within a partition, retention is time- or size-bounded, and consumers can rewind. Operations-heavy compared to RabbitMQ/NATS; worth the cost at >100k msg/s or when log replay drives the architecture. Alternatives: Redpanda (Kafka-wire-compatible, written in C++, simpler ops), Pulsar (segmented storage, more flexibility).
RabbitMQ is the mature AMQP broker — teams reach for it when they need rich routing (direct, topic, fanout, header exchanges), per-message acknowledgments, and clustering across data centers. For simpler single-topic work queues, lighter options (Redis streams, NATS) often win on ops. For ordered high-throughput logs, Kafka beats it. For "I want a job queue my Rails/Celery app can talk to," RabbitMQ is still the default.
Nginx is on the critical path for most self-hosted apps: it fronts your Node/Python/Rails service, holds the Let's Encrypt cert, applies rate limits, and caches static assets. Teams reach for it over Apache for the lower memory footprint and the simpler config language, and over Caddy when they need the ecosystem of modules and tutorials rather than Caddy's zero-config magic.
Mastodon (AGPL-3.0) is the fediverse implementation that got the fediverse mainstream attention. Each server (instance) is independent, moderated by its own admins, but federates with every other via ActivityPub. Timeline is chronological, no algorithmic reshuffle. Run your own instance to control moderation and keep your community's data. Ruby on Rails + Postgres + Redis + Sidekiq + optional ElasticSearch. Alternatives: GoToSocial (Go, lighter), Pleroma (small-instance focus), Misskey (richer UI, Japan-origin).
Vaultwarden (AGPL-3.0, formerly bitwarden_rs) implements the Bitwarden API in Rust with a tiny resource footprint. Works with every official Bitwarden client (web, browser extensions, mobile apps, CLI) — unlike the official server, happy on a Raspberry Pi with 128 MB RAM. Pick it when you want self-hosted password management without running Bitwarden''s MSSQL/dotnet stack. Does lack enterprise features (SCIM, event logs) that Business tier offers in official server. Alternatives: Padloc, Passbolt.
Grafana is the default observability dashboard: operators run it next to Prometheus (metrics) and Loki (logs) to see what their stack is doing. The free OSS build (AGPL-3.0) covers 90% of what Grafana Cloud offers; paid features sit at reporting, SLO, and alert-destination management. Most self-hosted stacks pair it with Promtail or Fluent Bit for log shipping.
Airflow (Apache-2.0) is the incumbent batch-pipeline scheduler. Pipelines are Python code (DAGs), operators cover every common system (S3, Postgres, Spark, k8s, HTTP), and the UI shows runs, logs, and reruns. Teams run it for ETL, ML training pipelines, nightly reports, data warehouse loads. Expect a Postgres metadata DB, a scheduler, a webserver, and either Celery or k8s workers. Alternatives: Prefect (modern Python-native), Dagster (asset-oriented), Argo Workflows (k8s-native), Temporal (durable execution for long-running workflows).
Home Assistant (Apache-2.0) is the OSI home automation standard. Pulls together Zigbee, Z-Wave, MQTT, Matter/Thread, and REST/cloud APIs for everything from smart bulbs to vacuums to solar inverters. Automations are YAML or a visual editor; UI is per-room dashboards. Local-first — devices work when the WAN is down. ~2500 integrations maintained by a huge community. Pick it over SmartThings/Alexa Routines when you want privacy and offline reliability.
Qdrant stores embeddings alongside structured payloads and lets you filter the ANN search by those payloads — "nearest to this vector where tenant_id = 42 and language = en." Teams pick it over Chroma for throughput and over Milvus for simpler ops. Agent authors reach for it as the long-term memory store.
NetBox (Apache-2.0) is the IPAM/DCIM everyone eventually standardizes on. Model your racks, devices, interfaces, cables, IP prefixes, VRFs, VLANs, circuits, power. Automation reads from the API to configure switches, generate DNS, drive BGP. UI is approachable; the schema is strict enough that it's actually trustworthy as a source of truth. Pick it over spreadsheets the moment the network has more than one admin or more than ~50 devices.
WireGuard (GPL-2.0) replaced OpenVPN and IPsec as the default self-host VPN choice around 2020. Simpler config (one short public key per peer), far smaller codebase (~4000 lines of kernel code), better performance. The linuxserver image bundles it with wg-easy or a plain config generator for setting up roaming peers. Pick it for site-to-site tunnels, remote-worker access, or point-to-point between homes. Alternatives: Tailscale (SaaS mesh on top of WireGuard, zero config), NetBird (OSS Tailscale alternative), ZeroTier.
Jackett proxies ~500 BitTorrent and Usenet indexers behind one Torznab/Newznab API so Radarr/Sonarr don't need per-site logic. You add your indexer credentials to Jackett once; the arr apps query Jackett, Jackett queries the indexer. Prowlarr is the newer alternative with better arr-side integration.
HAProxy is what teams graduate to when they need true layer-4 load balancing, per-backend health checks with retry policies, or sticky sessions beyond what nginx handles well. It's also the quiet workhorse inside many managed Kubernetes ingress controllers. For most sites nginx or a cloud load balancer is simpler; reach for HAProxy when you need per-request connection pooling, advanced ACLs, or sub-ms latency targets.
Metabase (AGPL-3.0 open-source edition, with a paid tier) is the "data visualization my product manager can use" default. Non-engineers write questions via a GUI, engineers write SQL when they need to, both live in the same dashboards. Alternatives: Superset (Apache-2.0, more flexible but harder to run), Redash (acquired by Databricks), Lightdash (dbt-native). Metabase shines when the audience is "the whole company."
Jellyfin (GPL-2.0, forked from Emby in 2018 when Emby closed source) is the fully-OSI media server. Point it at a folder of movies, TV, music, or books; it transcodes on demand and serves via apps for iOS, Android, Apple TV, Roku, LG/Samsung TVs, and web. Pick it over Plex when you want no cloud dependency and no premium-features paywall, and over Emby for the OSI licence. Pair with Radarr/Sonarr for automated library curation.
Fluent Bit (Apache-2.0, CNCF graduated) is the small-footprint log/trace/metric collector — C-written sibling to Fluentd. ~1 MB binary, memory-efficient enough for sidecars and edge devices. Reads from files, systemd, docker, kubernetes, HTTP, OpenTelemetry; parses JSON, regex, multiline; ships to Elasticsearch, Loki, S3, Kafka, Splunk, Datadog, CloudWatch, and many more. The default log collector in most modern k8s observability stacks (replaces Fluentd when size matters).
Keycloak (Apache-2.0, Red Hat) is the self-hosted identity provider when you want OIDC and SAML for everything (your internal apps, GitHub Enterprise, Jenkins, cloud consoles) and don''t want to pay Okta or Auth0 per MAU. Comes with user federation to LDAP/Active Directory, a themeable login screen, and an admin console for role/permission management. Alternatives: Authentik (Python, lighter), Zitadel (Rust, newer), Dex (minimal, Kubernetes-native).
Transmission is the lightweight BitTorrent client that pairs with Radarr/Sonarr when the file you want isn't on Usenet. Teams run it on a VPN-connected container with the web UI on a private port, let the arr apps drop torrents in, and let it seed in the background. qBittorrent is the heavier-featured alternative.
Immich (AGPL-3.0) is the project that finally made self-hosted photos competitive with Big Tech. Mobile apps (iOS + Android) auto-upload from your phone; a web UI gives timeline, albums, face recognition, OCR search, memories, maps. ML runs locally. Pick it over PhotoPrism (AGPL-3.0, more mature) when the mobile apps and face recognition matter most; PhotoPrism for a smaller resource footprint and polished indexing.
MariaDB is where you land if you want MySQL without Oracle, or if your distro picked it as the apt-installable default. Day-to-day it looks like MySQL — same drivers, same SQL — but the storage engines (Aria, ColumnStore) and the governance diverge. Pick it when existing tooling expects "the MySQL protocol" and you want the version with the clearer open-source commitment.
Bazarr watches your Plex/Jellyfin library and the arr apps for new content, then fetches subtitles from OpenSubtitles, Addic7ed, Subscene and friends. Can hash-match for perfect sync or OCR-fall-back when the hash misses. Pure-arr-stack companion; no standalone use.
Lidarr (GPL-3.0) is the music counterpart to Sonarr (TV) and Radarr (movies). Watch an artist''s discography on MusicBrainz; when a monitored album appears on a configured indexer (Usenet or torrent), grab it, rename, tag, hand off to your Plex/Jellyfin/Navidrome library. Weak link historically has been audio-metadata quality; most users pair it with Beets or picard for post-import cleanup. Alternatives: Headphones (older), Beets alone (CLI-only).
Neo4j (GPL-3.0 Community Edition, with an Enterprise tier above) stores graphs natively — relationships are first-class pointers, not JOIN tables. Teams reach for it when the data is a graph: fraud detection networks, recommendation systems, biology / knowledge graphs, dependency analysis. The Community Edition caps at single-node; Enterprise adds clustering and more granular security. Alternatives: Dgraph (newer, distributed), ArangoDB (multi-model), FalkorDB (Redis module).
CrowdSec (MIT) parses logs (nginx, SSH, cloudtrail) with user-contributed "scenarios", decides whether a source is hostile, and ships that verdict to bouncers at the edge (iptables, nginx, Cloudflare, Traefik). The novel piece is the Central API — anonymized attack signatures from the community feed back into every install's blocklist, like Project Honeypot for the modern web. Pick it over Fail2Ban when you want shared threat intel; keep Fail2Ban for air-gapped hosts.
Nextcloud (AGPL-3.0) is the self-hosted replacement for Google Workspace or Microsoft 365. Deploy once, get sync clients for every OS, server-side encryption, external storage mounts, and ~200 apps for everything from email to CAD viewing. Picks over Seafile or Syncthing when you want more than file sync; competes with ownCloud (the fork it came from) and OnlyOffice. Resource-heavy; plan for ≥4 GB RAM.
Airflow is where data teams land when cron jobs stop being enough: they need dependency graphs, retries, backfills, and a UI showing which runs failed. The DAGs are Python; the scheduler and worker are separate processes you orchestrate with Celery, Kubernetes, or the LocalExecutor. Alternatives: Prefect (newer, prettier UI), Dagster (type-aware, asset-first), plain cron with cronic (if the DAG is a straight line).
Syncthing (MPL-2.0) runs on each device, discovers peers over LAN or the shared relay network, and continuously syncs the folders you share. No central server, no account, no cloud. Teams run it alongside Nextcloud when they want sync without cloud dependencies; solo users run it to keep laptop/phone/desktop in step without Dropbox. The protocol handles conflicts via file versions, not merges.
CouchDB (Apache-2.0) stores JSON documents and offers one feature no other mainstream database matches: multi-master replication that handles conflicts cleanly. PouchDB is the browser counterpart — write locally, sync when online. Reach for it when the app runs partially-offline (medical devices, field data collection, mobile). For a general-purpose document store, MongoDB (SSPL — not CE) or Postgres' JSONB are the common defaults.
Woodpecker CI (Apache-2.0) is the community fork of Drone that chose to keep every feature OSS after Harness took Drone commercial. API-compatible with Drone pipelines for the common cases; smaller surface, faster to self-host. Pairs with Gitea/Forgejo and GitHub/GitLab via standard webhooks. Pick it over Drone for OSI-strict shops; stick with Drone when you''re on their hosted plan or need the enterprise-only pieces. Alternatives: Gitea Actions (GH-Actions-shaped, built into Gitea).
Jenkins (MIT) is the incumbent CI server for the past twenty years. Hundreds of plugins, Groovy-based pipelines, master/agent architecture, works with every SCM and every artifact store. Enterprise shops run it because it integrates with their existing toolchain. For greenfield work most teams now pick GitLab CI, GitHub Actions, Drone, or Woodpecker — simpler setup and newer UIs. Jenkins wins on "the plugin for that obscure thing exists".
qBittorrent (GPL-2.0) is the heavy-featured BitTorrent client — categorization, scheduling, bandwidth per-torrent, RSS auto-download, trackers whitelist. Self-hosters run it in a container (often behind a VPN sidecar via gluetun) and feed it from Radarr/Sonarr. Transmission is the lighter alternative when you only need "download and seed." Deluge if you want a client with a plugin ecosystem.
Jitsi Meet (Apache-2.0) is the OSI alternative to Zoom/Google Meet for ad-hoc rooms. Create a URL, share it, start a call. End-to-end encryption optional, lobby/password controls, recording via Jibri, moderation. Scales to roughly a few hundred participants per server with Videobridge/Octo. Pick it when privacy matters or you want to avoid per-user SaaS pricing. Alternatives: BigBlueButton (classroom-focused, heavier), Nextcloud Talk (bundled with Nextcloud).
OpenSearch (Apache-2.0) is where teams land after Elasticsearch moved to the Elastic License in 2021. Fork of Elasticsearch 7.10 with its own distribution team, security plugin bundled free, and OpenSearch Dashboards as the Kibana fork. API-compatible for the common cases; some newer Elasticsearch features (vectors, ESQL) have OpenSearch counterparts that landed later. Pick it over Elasticsearch for any new deployment where OSI matters.
Mattermost (AGPL-3.0 team edition, with Enterprise above) is the Slack-shaped self-hosted chat default. Teams pick it when "our conversations can't live in Slack" — regulated industries, air-gapped environments, cost-conscious orgs with >50 users. The Enterprise edition adds compliance exports, LDAP groups, and SSO integration beyond the AGPL build. Alternatives: Rocket.Chat (closer UX to Slack), Zulip (topic-threaded by design).
Memcached is deliberately smaller than Redis: no persistence, no pub/sub, no data structures beyond strings and integers. Teams run it for session storage, rendered-page caching, and database-query result caching. If you need lists, sorted sets, or anything that survives a restart, you want Redis (or Valkey) instead.
Grafana Agent (Apache-2.0) packages Prometheus-compatible scraping, Loki log tailing, Tempo trace shipping, and process-exporter metrics in a single binary. Reach for it when "one agent, three signals" matches your deployment model better than three separate agents. For single-signal use the dedicated agent (Promtail for logs, standalone Prometheus for metrics); for mixed signals the Agent is simpler.
InfluxDB (MIT for OSS v1/v2, with Enterprise and Cloud above) is the time-series DB used widely for IoT, observability, and real-time analytics. v2 bundled a UI and Flux query language; v3 (not yet fully open) builds on Apache DataFusion/Arrow for columnar storage. For a plain Prometheus-style metrics stack, Prometheus is the default; InfluxDB wins when you need event+metric mixed workloads and higher-cardinality tags. Alternatives: TimescaleDB (Postgres extension), QuestDB, VictoriaMetrics.
Coder (AGPL-3.0) runs development environments as managed workspaces: each workspace is a Terraform template (a VM, container, k8s pod) with your dotfiles, SSH keys, and a browser VS Code. Platform teams run it so engineers get consistent, powerful dev machines without juggling local setups. Pairs well with devcontainers. Alternatives: Gitpod (SaaS and self-hosted), GitHub Codespaces (SaaS only), DevPod (CLI-driven).