FAQ

Questions

  • Can AstraGo run on our existing GPU servers?

    Yes. AstraGo works with existing NVIDIA GPU servers and NPUs (FuriosaAI, DeepX), and heterogeneous GPUs can be managed together on a single platform. We analyze your current infrastructure and requirements, then design and build a GPU cluster structure that fits your environment.

  • Which GPU environments are supported?

    AstraGo supports the full NVIDIA GPU lineup and can be designed around on-premise environments. It supports both AI training and inference workloads, and GPU clusters built from multiple servers can be managed as one.

  • We want to use AstraGo in an air-gapped environment.

    AstraGo can be used in air-gapped environments and other settings where security matters. It can be deployed with your security and network requirements in mind, and supports operations suited to on-premise environments.

  • How do we get technical support after deployment?

    A dedicated developer is assigned at deployment and provides technical support directly. When an issue occurs, a dedicated engineer identifies the cause and resolves it, with remote or on-site technical support. We also provide the technical support needed for operations, including operator training, system configuration changes, and integration and application work.

  • We want to run AI training and inference workloads together.

    AstraGo supports both AI training and inference workloads. GPU resources can be partitioned and allocated to match each workload, optimizing utilization so that several workloads run on a single GPU at the same time.

  • Can we verify the benefits before adopting it?

    Yes. You can run a proof of concept (PoC) in your own environment to verify the benefits first. Details such as the PoC duration and scope are agreed separately.

  • How long does deployment take?

    It takes roughly two weeks to two months, depending on the level of customization, such as security, networking, storage, and integration with external systems.

  • How is pricing structured?

    Pricing consists of an annual subscription license and a per-node server installation fee. Custom development, if needed, is agreed separately.

  • Can it be installed on servers already in operation or in an existing Kubernetes environment?

    On servers already in operation, we install Kubernetes first and then AstraGo. It can also be installed in environments already running Kubernetes, but because it may conflict with existing software, our team checks in advance before installation.

  • How does power management work?

    While batch jobs such as training and fine-tuning are running, the GPU power cap is adjusted automatically. When the job finishes or the feature is stopped, GPU settings are automatically restored to their defaults.

  • Which NPUs are supported?

    AstraGo currently supports FuriosaAI and DeepX NPUs, which can be used for inference serving of vision models (object detection and image classification) from the DeepX ModelZoo.

  • Which AI tools can the AI Agent connect to?

    It connects via MCP and has been verified with Claude Code. Codex and ChatGPT are also supported.