Five new collections: where to start if you don't know what to track
Published: 8/1/2026
The hardest part of a tool that tracks repositories is the empty state. You know GitGenius watches thousands of projects. You do not know which of them you care about, and "search for a repo" assumes you already have a name in mind.
Collections are the answer to that, and we just added five: Cloud Native & Kubernetes, Observability & Monitoring, API & Microservices, Security & Identity, and Developer Tools & Automation. They join AI, Machine Learning, Containers, Databases, and ELT.
Cloud Native & Kubernetes
Orchestration, deployment tooling, and the platform layer underneath it. Assembled from repos tagged kubernetes, cloud-native, container orchestration, orchestration, deployment, and openshift.
A few that land here: Kong/kong, the API gateway; GoogleContainerTools/skaffold and GoogleContainerTools/kaniko for build-and-deploy loops; GoogleCloudPlatform/microservices-demo, which is what half the industry actually learns Kubernetes from.
Observability & Monitoring
Metrics, logging, tracing, and the tooling that tells you what production is doing. Built from observability, monitoring, metrics, logging, and performance monitoring.
VictoriaMetrics/VictoriaMetrics and SigNoz/signoz are here, along with GoogleChrome/web-vitals — front-end performance measurement counts as observability, even if it rarely gets filed that way — and PrefectHQ/prefect on the orchestration-visibility side.
API & Microservices
Gateways, frameworks, service architecture, and the plumbing between services. From api, apis, api integration, api gateway, microservices, middleware, and routing.
Kludex/starlette underneath most modern Python APIs, IBM/sarama for Kafka in Go, and AykutSarac/jsoncrack.com, which is the thing you open when an API response makes no sense.
Security & Identity
Authentication, authorization, policy enforcement, and hardening. From security, authentication, and policy enforcement.
FiloSottile/mkcert for local certificates that actually work, apache/casbin for authorization models, OpenZeppelin/openzeppelin-contracts on the contract side, and CISOfy/lynis for system auditing.
This is the narrowest collection of the five, and deliberately so. "Security" is a category that will absorb anything you let it.
Developer Tools & Automation
The largest of the new collections, covering CI/CD, code generation, testing, linting, and workflow automation — developer tools, tooling, automation, workflow automation, ci/cd, continuous integration, and code generation.
Aider-AI/aider, BerriAI/litellm, ComposioHQ/composio, and GoogleCloudPlatform/terraformer all live here. The AI-assisted-development wave has landed squarely in this collection over the past year, and it shows.
How these are actually assembled, and where that leaks
Each repository we index carries derived category tags. A collection is a list of tags plus everything matching any of them, deduplicated and sorted alphabetically by repository name.
Three consequences worth being upfront about.
Repos appear in more than one collection. Kong shows up in cloud native, observability, APIs, *and* security, because it genuinely is all four. That's not a bug in the tagging; an API gateway that does rate limiting and auth belongs in every one of those lists.
The ordering is alphabetical, not ranked. A collection is not a leaderboard and the position of a repo in it means nothing. If you want ranking, the repo pages have the actual numbers — response times, issue throughput, star growth — and those are the comparison worth making.
The tags are derived, so they're imperfect. Something you'd expect in a collection will occasionally be missing from it. If you spot a repo that's obviously miscategorized, tell us — that feedback is more useful than it sounds, because the failure is usually systematic rather than one-off.
Where to start
Pick the collection closest to what you work on and open two or three repos you already depend on. The interesting read isn't the project you know is healthy — it's the one you assumed was, seeing its median first-response time for the first time.