diff --git a/content/blog/2026-10-02-modelplane-v0-5/index.mdx b/content/blog/2026-10-02-modelplane-v0-5/index.mdx new file mode 100644 index 0000000..cb585aa --- /dev/null +++ b/content/blog/2026-10-02-modelplane-v0-5/index.mdx @@ -0,0 +1,183 @@ +--- +title: "Modelplane v0.5: an AI gateway and fleet telemetry" +description: "Modelplane v0.5 turns the fleet gateway into an AI gateway and collects every engine's metrics under one vocabulary. Civo joins the clouds it can provision on." +date: "2026-10-02" +authors: + - name: "Dennis Ramdass" + title: "Principal AI Engineer, Upbound" + url: "https://github.com/dennis-upbound" + avatar: "/authors/dennis.jpg" + github: "https://github.com/dennis-upbound" + bio: "Dennis is a Principal AI engineer at Upbound and a core maintainer of Modelplane. He's spent the last two decades working on cloud and infrastructure, and more recently AI agents and infrastructure, and is now bringing that work to AI inference with Modelplane." +tags: ["release", "inference", "control-plane", "observability"] +draft: true +pinned: true +--- + +Modelplane v0.5 is out. The gateway on your control plane is now an AI +gateway: it authenticates callers, reads the model a request asks for, and +fails over between the backends that serve it. Modelplane also collects your +fleet's metrics under one set of names, whatever engine produced them. And +Civo joins the clouds Modelplane can provision a cluster on. + +A `ModelDeployment` can also scale to zero replicas now, so one you aren't +serving from costs you no GPUs. Here's what's new. + +## The fleet gateway is an AI gateway + +The gateway on the control plane used to be an HTTP router. It understood +nothing about the requests it forwarded. A caller reached a `ModelService` by +path prefix, nothing authenticated them, and the hop out to each cluster +crossed the public internet in plain HTTP. + +It's now an [Envoy AI Gateway](https://aigateway.envoyproxy.io/), the same one +every `InferenceCluster` already runs at its edge. It reads the model a request +names in its body and resolves the `ModelService` that serves it. For the +backend it picks, it rewrites the model name, the credential and the path, so a +backend sees the name it knows and a caller's key never reaches a third party. + +`ModelService` gains `priority` alongside `weight`. Weight splits traffic +between backends at one priority; priority fails over to the next when they go +unhealthy. Every request meters a token count per caller, streams included. + +`InferenceGateway` also stops being a singleton. It names the cluster it runs +on, so you can run one per region for residency, or two in a region for +availability: + +```yaml +apiVersion: modelplane.ai/v1alpha1 +kind: InferenceGateway +metadata: + name: eu +spec: + clusterName: gw-gcp-eu + tls: + certificateRefs: + - name: eu-example-com-tls + auth: + method: APIKey + apiKey: + secretSelector: + matchLabels: + modelplane.ai/inference-keys: "true" + serviceSelector: + matchLabels: + example.org/region: eu +``` + +The hop from a fleet gateway to a cluster gateway is now authenticated in both +directions by a per-cluster PKI, which cert-manager issues and trust-manager +distributes. + +## One vocabulary for a fleet's metrics + +Modelplane doesn't own your engine. You bring the image and the command, and +that's the point: a `ModelDeployment` runs vLLM, SGLang, or anything else that +speaks the OpenAI API, without Modelplane knowing anything about it. + +That same freedom is what makes a fleet hard to watch. vLLM publishes +`vllm:num_requests_waiting`. SGLang calls the same measurement +`sglang:num_queue_reqs`. DCGM reports framebuffer memory in mebibytes under a +name that says bytes, and energy in millijoules under a name that says joules. +A dashboard written against one engine is wrong on the next, and a fleet +running both has no fleet-wide number at all. We couldn't fix that by picking +an engine, so we fixed it at collection. + +Modelplane now runs an OpenTelemetry collector on every inference cluster. It +discovers every component Modelplane installs, renames each one's series into a +single `modelplane_*` vocabulary, and exports them wherever you say. Only +`modelplane_*` leaves the cluster: a series nobody renamed is one whose meaning +Modelplane can't vouch for across engines, and it costs the same to carry as +one that was. + +Two new kinds. A `TelemetryDestination` says where metrics go, and nothing is +collected until one exists: + +```yaml +apiVersion: modelplane.ai/v1alpha1 +kind: TelemetryDestination +metadata: + name: default +spec: + sinks: + - name: prometheus + type: prometheus_remote_write + endpoint: https://prom.example.internal/api/v1/write +``` + +A `MetricMapping` says what a component emits and what Modelplane calls it. +Modelplane comes with mappings for vLLM, SGLang, the gateway, the endpoint +picker and DCGM, so those need nothing from you. Write one for an engine Modelplane +has never seen and its numbers join the same surface: + +```yaml +apiVersion: modelplane.ai/v1alpha1 +kind: MetricMapping +metadata: + name: my-engine +spec: + metrics: + - from: my_engine_queued_requests + to: modelplane_requests_waiting + - from: my_engine_kv_transfer_ms + to: modelplane_request_kv_transfer_seconds + fromUnit: Milliseconds +``` + +`fromUnit` is worth explaining, because it encodes something a dashboard would +otherwise have to guess. A metric's name is no guide to its unit, so the mapping +says what the source measures in and Modelplane converts to the base unit the +target name claims, histogram buckets and all. Skipping it is the expensive +mistake: a series named `_seconds` holding milliseconds reads a thousand times +fast, and nothing downstream can tell. + +One thing Modelplane deliberately doesn't do is combine a deployment's replicas +for you. Each publishes its own series, and a scrape of one replica is one batch, +so a collector that added them up would be summing readings taken at different +moments. That isn't the traffic that happened. Your backend already has every +replica's series and can combine them at query time. + +## Civo + +[Civo](https://www.civo.com/) joins EKS, AKS, GKE, Nebius and Vultr as a cloud +Modelplane can provision an `InferenceCluster` on, with the same spec you'd +write for any of them. + +Two things about Civo needed handling underneath. Its GPU images carry no +NVIDIA driver, so the serving stack installs the GPU Operator to supply one, +with the toolkit and device plugin switched off so the DRA driver stays the +only thing allocating GPUs. And Civo has no server-side autoscaler, so a pool +with a `maxNodeCount` is scaled by the upstream cluster-autoscaler running on +the cluster itself. Civo's volumes are ReadWriteOnce, so `ModelCache` isn't +available there yet. + +## Scaling to zero + +A `ModelDeployment` can now scale to zero replicas. `spec.replicas` used to carry +a floor of one, so `kubectl scale --replicas=0` was rejected at admission. That +is awkward when scaling to zero is most of the reason to put KEDA in front of a +GPU workload in the first place. There was no way to park a deployment +either: withdrawing its endpoints while keeping the object meant tainting the +cluster hosting it. + +Dropping the floor on its own would have made a parked deployment look broken. +Zero desired replicas against an empty schedule reads as none of them scheduled, +and on a control plane with no clusters it reads as having nowhere to run, so a +deployment you had deliberately parked would sit there permanently not ready. + +Zero now takes a path of its own. Nothing is composed, so the deployment's +`ModelReplicas` and `ModelEndpoints` are pruned, `status.replicas` reports 0 to +the scale subresource, and readiness reports true with a `ScaledToZero` reason, +the way a Deployment at zero replicas still reports Available. A parked +deployment reads as parked rather than as failing, which is what makes it safe +for an autoscaler to do on your behalf. + +## Try it + +The [getting-started guide](https://docs.modelplane.ai/getting-started/) covers +standing up a fleet, and +[Monitor the Fleet](https://docs.modelplane.ai/platform/telemetry/) covers +pointing telemetry at a backend you already run. Modelplane is Apache 2.0 and +moving fast at +[github.com/modelplaneai/modelplane](https://github.com/modelplaneai/modelplane), +and questions are welcome in [Slack](https://slack.modelplane.ai).