Modern CI/CD You Cannot Miss for AI-Powered Deployments
Ring Promoter is Workstation’s production-grade promotion control plane. It moves versions of many applications through ordered deployment rings — int → test → acc → prod — with live health gates, version-verified checks, automatic rollback, and a full audit trail. Built for teams shipping AI platforms, agents, and APIs at pace.
A small control plane that promotes application versions through shared rings. It runs as a long-lived service (typically on k3s), exposes a JSON REST API and an embedded web UI, and tracks current/previous versions and history per (application, ring). Kubernetes apps and VM/CI apps share one surface — the deployer is chosen per app.
Every release earns production. One hop at a time — never skip a ring — with health before promote and auto-rollback on failure.
First landing for new builds and AI agent / API changes.
Automated verification and soak before acceptance.
Stakeholder / UAT gate — often the last auto-promote stop.
Live traffic — password-gated for the final ring when configured.
Start the pipeline by seeding a version (image tag, branch, or ref) into the Integration ring for an application.
promote {from_ring} always targets the next ring. You cannot jump int → prod. The source ring must be healthy first.
KubectlDeployer (set image + rollout), GitHubActionsDeployer (workflow-dispatch for VM/CI apps), or k8sjob runner Jobs — selected per application.
After deploy, HTTP health checks run with configurable retries. Optional version verification requires the endpoint to report the exact version just deployed.
If the target stays unhealthy, Ring Promoter rolls back to the previous version automatically and records the failure in history.
Flag a ring so a healthy landing continues onward in the same locked operation — e.g. auto-promote test → acc, but keep humans on the gate to prod.
AI products change fast — models, agents, MCP tools, gateways. Ring Promoter keeps promotions disciplined without freezing delivery.
Shared pipeline int → test → acc → prod. Adding a ring is a single-place change in the ring package — apps declare where they live per ring in config, not code.
Live health checks before and after promote. Optional health_version_field / header stops the classic false green: old pods answering 200 while the new version never rolled.
Failed health after retries restores the previous version. Seed, promote, and rollback are all audited — success or failure.
One control plane for kubectl rollouts, GitHub Actions workflow-dispatch (e.g. WSL Proxy on VMs), and Kubernetes Job executors with live logs.
Per-app serialization via Postgres session advisory locks across replicas. Operations detach from HTTP so LB timeouts cannot abort a rollback mid-flight.
Embedded Next.js console and bearer-token REST API. Async deployment views stream progress for operators watching AI stack releases.
Add an application with YAML: namespaces, Deployments, image repos, health URLs, deployer type — no fork of Ring Promoter required.
Promote agent services, MCP gateways, GPU inference front-doors, and Workstation products through the same rings your platform team already trusts.
Model weights, agent runtimes, and edge gateways ship on different cadences. Ring Promoter keeps every change on the same promotion protocol.
Clean interfaces: Deployer, Executor, Health Checker, Store — swap production backends without rewriting promotion rules.
Go service + embedded UI + REST API on k3s (or local).
Ordered pipeline defined once; apps map resources per ring.
kubectl, GitHub Actions, or k8sjob executors per app.
Postgres (prod) or memory (dev) with history and locks.
Explore the product on ringpromoter.com, read the Workstation deep dive, or talk to us about CI/CD for AI platforms.
Visit https://www.ringpromoter.com/ for demos, docs, and the latest Ring Promoter releases.