DOCSEARCHbyAlgolia
Submit Docs

Hyperstack Documentation

website·Indexed 34 days ago

https://docs.hyperstack.cloud

Documentation for Hyperstack's cloud platform, featuring guides on deploying virtual machines, managing GPU hardware and flavors, and orchestrating Kubernetes clusters. Includes technical details on confidential computing, spot VMs, CSI drivers, and enhanced monitoring for NVIDIA GPUs and AI inference workloads.

Tokens
235.9K
Snippets
864
Records
1.9K
Agent score
62%

What's inside Hyperstack

  1. Overview of Hyperstack Object Storage API

    Hyperstack provides an S3-compatible Object Storage service. The API allows developers to manage the infrastructure surrounding the storage, specifically inspecting buckets and managing the access keys used to authenticate SDKs against the storage service.

  2. Overview of Hyperstack Virtual Machine capabilities

    Hyperstack provides management for GPU and CPU virtual machines. Key capabilities include:

    • Lifecycle Management: Starting, stopping, rebooting, and monitoring VM states.
    • Configuration: Using Cloud-Init scripts for launch configuration and Provisioning Profiles to save configurations for future deployments.
    • Storage & Recovery: Creating snapshots for backups and building custom OS images from those snapshots.
    • Cost Optimization: Utilizing Spot Virtual Machines for reduced-cost GPU workloads and Hibernation to preserve state while reducing costs.
    • Monitoring: Accessing performance metrics (CPU, memory, GPU), event logs, console logs for diagnostics, and Enhanced Monitoring via Grafana.
    • Hardware Scaling: Changing VM flavors by recreating the instance from a snapshot.

  3. Overview of Hyperstack AI Studio capabilities

    Hyperstack AI Studio provides inference for a catalog of text and image models. Key features include:

    • OpenAI-compatible API: Send requests to models using standard API integrations.
    • Interactive Playgrounds: Test and tune model parameters with side-by-side comparisons.
    • Knowledge Bases: Index custom documents to ground model answers and provide citations.
    • Dedicated Inference: Deploy open-weight models on reserved GPUs for private, API key-protected endpoints.

  4. Overview of Hyperstack Network Firewalls

    Hyperstack network firewalls filter inbound (ingress) and outbound (egress) traffic for virtual machines. They allow or deny access based on direction, source/destination IP, protocol, ethertype, and port range. Firewalls can be implemented as reusable firewall objects attached to multiple VMs or as direct firewall rules applied to a specific VM.

  5. Overview of Hyperstack High-Performance File Storage

    Hyperstack's High-Performance File Storage is designed for complex, data-intensive workloads including High-Performance Computing (HPC), big data analytics, and AI model training. The system supports shared file systems with snapshot capabilities for data backup and recovery. Currently, the solution is in the development stage and supports Ubuntu 22.04.

  6. Overview of Public and Private IP addresses in Hyperstack

    Hyperstack uses two types of IP addresses for virtual machines (VMs):

    • Public IP Addresses: Enable VMs to communicate with external networks over the internet. They are essential for web hosting, API responses, and SSH access. Public IPs can be dynamic (may change when the VM is stopped/deallocated) or static (consistent for DNS or firewall rules). Public IPs are billed hourly.
    • Private IP Addresses: Confined to the internal network of a Virtual Private Cloud (VPC). They allow internal communication between resources (e.g., a backend database) without exposure to the public internet and persist for the VM's lifecycle within the VPC.

  7. Overview of Dedicated Inference

    Dedicated inference allows you to deploy a single open-weight model on GPUs reserved exclusively for your organization. Unlike shared endpoints billed per token, dedicated inference is billed by the hour for as long as the machine runs. It provides predictable latency, a private HTTPS endpoint protected by an API key, and the ability to run models not available in the shared catalog.

  8. Overview of Hyperstack AI Studio capabilities

    Hyperstack AI Studio allows users to run text and image models via a console UI or an OpenAI-compatible API. Key capabilities include:

    • Text Playground: Chatting with hosted models.
    • Image Generation: Creating or editing images from text prompts.
    • API Access: Programmatic interaction via an OpenAI-compatible interface.
    • Knowledge Bases: Grounding model answers in custom documents.
    • Integrations: Connecting AI Studio to tools like Cursor, Claude Code, and n8n.

  10. Overview of Hyperstack Pricebook API

    The Pricebook API allows developers to manage and forecast cloud spend by providing access to current pricing rates for all resource types. It specifically handles standard pricing as well as long-term GPU reservation contracts that override standard rates. Key capabilities include forecasting spend, looking up rates for running resources, and inspecting active contract terms.

  11. Overview of Enhanced Monitoring modes

    Hyperstack offers two monitoring modes for virtual machines. Standard mode provides basic metrics (CPU, RAM, and coarse disk I/O) without requiring an agent. Enhanced mode installs a lightweight agent inside the VM to collect high-granularity data, including GPU utilization, temperature, power draw, fan speed, RAM, and finer disk I/O with longer retention. The agent is open source and available in the hyperstack-agent repository.

  12. Overview of Large Root Disk Flavors

    Large Root Disk Flavors are VM configurations that eliminate the separate ephemeral disk and reallocate its capacity to the root disk. This provides a single, larger persistent storage device for both the operating system and workloads. These flavors are identified by the -bigroot suffix in their name.

    Key Characteristics:

    • Storage: Single large root disk; no ephemeral disk (ephemeral: 0).
    • Persistence: Root disk is persistent for the life of the VM but is local-only and cannot be recovered once the VM is deleted.
    • Availability: Available via API only; they are not visible in the Hyperstack UI for selection.
    • Restrictions: They do not support hibernation or snapshotting.
    • Billing: Because hibernation is unsupported, stopping the VM does not pause billing. The VM must be deleted to stop charges.