// the find
pipe-cd/pipecd
The One CD for All {applications, platforms, operations}
PipeCD is a GitOps-based CD platform that tries to handle Kubernetes, Terraform, Lambda, ECS, and Cloud Run through a single control plane and a unified pipeline definition format. The architecture splits into a central control plane and per-cluster 'Piped' agents that run close to the target environment, so deployment credentials never leave your cluster. It's aimed at platform teams managing deployments across multiple clouds or environments from one place.
The agent model is architecturally sound — Piped pulls work from the control plane rather than the control plane pushing into clusters, which is the right call for credential isolation and firewall-friendliness. Built-in deployment analysis (metrics, logs, request sampling) as a first-class pipeline stage is genuinely useful and not something you get out of Argo CD without extra tooling. The v1 plugin architecture is a meaningful improvement over the monolithic v0 approach, making it possible to add platform support without forking the core. Active CNCF project with regular releases and a real adopter list — not vaporware.
Still CNCF Sandbox with ~1300 stars, which means the community is small and you're likely to hit edge cases with limited help. The v0-to-v1 migration path exists but the fact that they're running both in parallel (pipedv1 as a separate binary) signals the plugin architecture isn't fully baked yet — adopting v1 now means being an early tester. Control plane requires its own persistent infrastructure (runs on K8s or ECS), so the operational footprint is heavier than tools like Flux that piggyback on the cluster you're already managing. Documentation covers the happy path but the configuration reference is dense and the difference between 'platform provider' and 'analysis provider' isn't immediately obvious to someone onboarding alone.