Skip to main content

Converge

A declarative orchestration engine for platform teams. Describe the resources you want as specs; the engine reconciles the world to match — composing, ordering, healing, and rolling up status.

Built for scale (millions of resources), simplicity (a single binary + Postgres), and reactivity (work is pushed, not polled).

Get startedExplore the demosGitHub
Converge converging a resource graph in the UI

Describe what you want. Converge makes it so.

Like Crossplane or kro, it composes a resource into a graph and rolls readiness up — but dependency edges and value flows are declarative data you can edit live, in the language of your choice.

One spec fans out into a whole graph

Declare a high-level resource once; Converge expands it into a graph of children, brings them up in dependency order, and flows each one's outputs — an ID, an ARN, a URL — into the specs that depend on it. A single landingzone spec becomes a hundred accounts, each with its VPCs, routes, and guardrails, wired together with no glue code.

Never drifts, never lies about status

Every resource is continuously reconciled — anything that drifts is detected and healed, across millions of resources. A composite is green only when everything beneath it is healthy, and when it isn't, it names the exact child at fault — status you can actually trust.

Logic is data — edit it live, no redeploy

Composition rules, dependency edges, and value flows are first-class declarative data, not controller code. Write them in Go, CEL, or Starlark — or ship them in a config bundle and edit them on a running cluster. Change how a kind composes and every resource re-converges. No rebuild, no rollout.

Provider-agnostic core

The engine knows no concrete resource type. Teach it a kind with a Kubernetes-CRD-style manifest and a handler that does anything — Terraform, a Kubernetes Job, or any cloud API call.

Any language — or none

Write providers in Go or TypeScript, or make any external program a kind over stdio — a bash script, a Python file, a binary — with no SDK import or compile step.

One binary, one database

No etcd, Redis, Kafka, or message broker. A single binary over Postgres does it all — leaderless, self-sharding, and colocated or split into control and worker tiers as you grow.

Reactive at scale

Thousands of reconciliations per second, tuned for millions of resources. Work is pushed, not polled, so a change propagates immediately — and at-least-once, idempotent delivery makes it safe to redeliver.

Live lifecycle & migrations

Version a kind and migrate a resource live (vpc/v1 → vpc/v2). Throttle a kind cluster-wide with concurrency caps, roll back to any earlier spec, and fire durable follow-up reactors on any transition.

GitOps, CLI & live UI

conctl sync reconciles a manifest directory Argo-CD-style with ownership tracking and prune. A kubectl-style CLI and a rich web UI — topology, dependency graph, fleet view, live events — ship in the box.

See every feature at a glance →