For teams building AI infrastructure products.

Your AI product.Our foundation.

Build your own model-serving, training or agent product. Astro Platform runs its supported workloads in infrastructure you control, from Console, CLI, API, or Nova with approved access.

Start freeTalk to sales
Explore AI Infrastructure examples

Inference

Requests → router → serving workers

APIRequestsGPU 01Model servingGPU 02Model servingOPERATIONSDeploy · scale · observeROUTERYour routing logic

Who it is for

Build the part only your team can build.

  • You have a runtime, scheduler, inference engine or data system that makes your product different. The production foundation around it should be reusable.
  • You need to deploy and operate workloads across customer accounts or your own cloud without building a second platform product.
  • You need networking, lifecycle, organization access, cost visibility and operational controls without giving up your product IP.

Not yet

If you are calling a hosted model API and run no infrastructure of your own, you do not need this yet. The reason to come back is usually the invoice, the security review, or the first customer who asks where their data goes.

Build a business on it

Every customer is one more account. Not one more platform.

Run your product in each customer’s cloud, from one control plane.

Three clouds, any cluster

EKS · GKE · AKS · or register what you already run

Access you control

Organizations · roles · API keys

One public API

Console · CLI · SDKs · MCP · Nova

Built the same way: Run production AI inference

A worked example: a model API on your own domain in your AWS account, built on the same public API.

Read the example

Your environment

Your workloads stay in your cloud.

The agent connects out. Deployed workloads and model weights run in your infrastructure; desired state and operational telemetry cross the connection. Nova may send your questions and selected context to configured AI providers.

desired state

health · metrics · cost · findings

Reusable production capabilities

Your product. Your blueprint.

Model serving is the blueprint most teams start with. The same bricks compose the rest.

Deploy a model endpoint in your own cloud.

Use a managed serving path or bring serving software. AstroPulse supplies deployment, GPU infrastructure, endpoint access and supported node scaling, not your inference optimization.

Reference pattern · use Console, CLI, SDKs, API or Nova

Architecture · 01 / Your components

Hugging Face model or your weights
vLLM or your runtime
Your application

02 / AstroPulse infrastructure bricks

Hugging Face

Supported model artifact sources

GPU nodes

GPU node groups · scale-down subject to minimums

Model serving

Serving add-on · runtime and gateway configuration

Queueing + metrics

Workload queueing and infrastructure visibility

03 / Your environment

Endpoint · configured runtime and access policy

Build it. Keep it running.

Operations are part of the blueprint.

Use infrastructure health and available telemetry to investigate and review supported changes with Nova. Inference-specific performance tuning stays with your runtime.

Health

Ready · restarts · rollout state

Metrics

Workload health and available metrics

Cost

Infrastructure cost visibility

Security

Workload and image findings where available

Investigate with Nova

Events, history and metrics for the model. Logs only with your permission.

Review the plan

The cause with its confidence, the change, and the rollback it keeps.

Act and observe

Review a supported proposed change, approve it when appropriate and verify the actual application state.

Assemble with Nova

Nova can help compose the foundation.

Nova can help assemble and review infrastructure capabilities using connected skills. The Console, CLI, API and SDKs can also drive supported operations.

Serve Qwen3-8B on prod-inference behind API keys, under acme.ai.

Catalog for prod-inference: model-serving and gateway both need cert-manager, which is not installed yet.

Blueprint: review dependencies, gateway access policy and model-serving configuration, then run a dry run before applying.

01cert-manager02gateway03model-serving04qwen-chat

Plan ready · dry run before applying

Before you ask

The questions that come next.

Where do our workloads and data run?

Your deployed workloads, model weights and serving traffic run in infrastructure you control. AstroPulse also processes operational metadata needed to provide its services; using Nova may send prompts or selected context to configured AI providers. See our Security and Privacy pages for the data-handling details.

Where can models come from?

Hugging Face, S3, Google Cloud Storage, Azure Blob, a volume or an OCI registry, with your own credentials for private models.

Do we need Kubernetes expertise on the team?

You can get started from Console without managing Kubernetes directly. Astro Platform provisions supported clusters or connects existing ones, but your team still owns workload-specific configuration and production decisions. CLI, API and Nova are additional interfaces, subject to access.

Do idle GPUs keep costing us?

GPU node groups can scale down when unused, subject to provider, workload, and minimum-capacity settings. A model configured with a warm replica still incurs compute costs. Check the scaling configuration and available idle-capacity behavior for your chosen runtime.

Can we bring our own serving runtime?

The platform has a managed model-serving path and supports deploying your own serving software through supported application sources. Your inference engine, scheduling, routing, batching, KV-cache management and latency optimization remain yours. Infrastructure deployment and scaling are not inference-performance optimization.

What happens if we stop using AstroPulse?

Your workloads and clusters remain in infrastructure you control. If you leave, you may need to replace AstroPulse-specific management, deployment workflows and policies, and export operational records you want to retain.

What does it cost?

Astro Platform has a free Developer tier with no credit card required. Enterprise pricing is custom; talk to sales about your team and rollout. Your cloud provider bills its infrastructure usage separately. Nova access is free by request, with usage limits.

AI Infrastructure

What are you building?

Bring your runtime, serving stack or data infrastructure. Keep the differentiated parts; use Astro Platform for the machinery required to run them.

Start freeTalk to sales

WHAT YOU CAN EXPECT

  1. 01

    Keep your runtime and IP

    Your scheduler, serving engine, router, models and product behavior stay yours.

  2. 02

    Build in your cloud

    Compose compute, networking, deployment and lifecycle capabilities in infrastructure you control.

  3. 03

    Operate with guardrails

    Use Astro Platform for production controls and Nova for optional engineering assistance.