# Nirvana Labs

> Nirvana Labs is a performance cloud built around high-IOPS block storage. Adaptive Block Storage delivers 20,000 baseline IOPS with up to 600,000 burst at 0.3–1.5 ms latency, with no burst credits and no degradation over time. Compute, networking and managed Kubernetes are built on the same metal.

Canonical HTML: https://nirvanalabs.io/

This document is where the product summary, the fit guidance below and the FAQ live; the HTML page does not render them. Fetch the canonical URL for what it adds instead — benchmark results, case studies and the pricing calculator.

## When to use this

- Provisioning virtual machines, block storage volumes, VPCs and firewall rules programmatically, via API, Terraform, CLI or SDK.
- Running IOPS-bound or latency-sensitive workloads — vector databases, OLAP and OLTP databases, blockchain nodes, agent checkpointing — where sustained throughput matters more than burst credits.
- Standing up managed Kubernetes (NKS) clusters with dedicated controller nodes and customizable worker pools.
- Running blockchain RPC and archive nodes without operating the underlying infrastructure.
- Executing untrusted or agent-generated code in isolated Agent Sandboxes.

## When to look elsewhere

- Serverless functions or platform-as-a-service deploys — Nirvana Labs provisions infrastructure, it does not host application runtimes.
- Object storage or CDN. The storage product is block storage attached to compute.
- Managed GPU or hosted LLM inference. Published instance types are CPU-only AMD EPYC shapes, and the API exposes no GPU, accelerator or inference resource. Nirvana Labs runs the infrastructure around a model — storage, checkpointing, sandboxes — not the model itself.

## Common questions

### What is Nirvana Labs?

Nirvana Labs is a performance cloud built around high-IOPS block storage. Compute, networking and managed Kubernetes run on the same metal, so latency-sensitive and data-heavy systems get predictable throughput instead of the burst-credit model most clouds use.

### What IOPS and latency does Adaptive Block Storage deliver?

Adaptive Block Storage delivers 20,000 baseline IOPS with up to 600,000 burst, at 0.3 to 1.5 ms latency. There are no burst credits to exhaust and no degradation over time, so a volume performs the same in month twelve as it did on day one.

### Which workloads is Nirvana Labs built for?

IOPS-bound and latency-sensitive systems: vector databases, OLAP and OLTP databases, blockchain RPC and archive nodes, agent checkpointing, and managed Kubernetes workloads. It suits any workload where sustained throughput matters more than peak burst.

### What is Nirvana Labs not the right choice for?

It provisions infrastructure rather than hosting application runtimes, so it is not a serverless or platform-as-a-service target. It also does not offer object storage, a CDN, GPUs or hosted model inference — the storage product is block storage attached to compute.

### Does Nirvana Labs offer GPUs or managed LLM inference?

No. Every published instance type is a CPU-only AMD EPYC 9354P shape, and the API exposes no GPU, accelerator or inference resource. Nirvana Labs is the infrastructure around a model rather than the model itself: high-IOPS storage for vector databases and agent checkpointing, isolated Agent Sandboxes for agent-generated code, and managed Kubernetes to run the rest of the stack. Model inference belongs with a provider that hosts it.

### How do I provision resources?

Through the REST API, the Homebrew-installable CLI, the Terraform provider, the Go and TypeScript SDKs, or the web dashboard. The API is described in an OpenAPI 3.1 document, and the dashboard emits equivalent Go, TypeScript, curl and Terraform for any action you take in it.

### Can an AI agent operate Nirvana Labs directly?

Yes. An official Model Context Protocol server, @nirvana-labs/nirvana-mcp, exposes resource management as tools, and the OpenAPI document can be turned into function-calling definitions. API keys carry scoped permissions — read or edit, per resource type and per project — so an agent can be granted least privilege rather than full account access.

### How is Nirvana Labs priced?

Usage-based, billed per hour with no minimum commitment. Compute is $0.016 per vCPU-hour and $0.004 per GB of RAM per hour, block storage is $0.00013 per GB-hour, and egress is $0.02 per GB. A billing month is 730 hours, and annual commitments take 20% off.

### What uptime is committed in the SLA?

A quarterly uptime target of 99.9% for managed RPC node APIs and Nirvana Cloud. If uptime falls between 99.00% and 99.9% the service credit is 10% of the affected quarterly spend, rising to 20% below 99.00%, capped at 90 days of paid service per quarter and claimable within 30 days. Standard support responds in under 24 business hours.

### What does the managed Kubernetes service include?

Nirvana Kubernetes Service (NKS) provisions clusters with dedicated controller nodes and customizable worker pools, plus managed load balancers and persistent volume claims backed by the same block storage. An official Terraform module covers production-ready cluster setups.

### How does performance compare to the major cloud providers?

Benchmarks against AWS, GCP and other providers are published with their full methodology and raw numbers, covering storage throughput, LangChain and LangGraph checkpointing, agent sandboxes and the Star Schema Benchmark, so the results can be reproduced rather than taken on trust.

## Developer resources

- [API & product documentation](https://docs.nirvanalabs.io/): Reference for every resource, endpoint and authentication flow.
- [CLI (Homebrew)](https://github.com/nirvana-labs/nirvana-cli): Official command-line client for VMs, networking, storage and RPC nodes. Supports pipes, pagination flags and shell completions.
- [MCP server](https://www.npmjs.com/package/@nirvana-labs/nirvana-mcp): Model Context Protocol server — the most direct way for an agent to drive Nirvana Labs resources.
- [Terraform provider](https://registry.terraform.io/providers/nirvana-labs/nirvana/latest): Declarative provisioning for all resource types.
- [Terraform NKS module](https://github.com/nirvana-labs/terraform-nirvana-nks): Opinionated module for production-ready managed Kubernetes clusters.
- [Source & SDKs](https://github.com/nirvana-labs): GitHub organization hosting the Go SDK, TypeScript SDK, CLI and Terraform provider.
- [Runnable examples](https://github.com/nirvana-labs-examples): End-to-end reference deployments, including LangChain and LangGraph benchmarks.
- [FAQ](https://docs.nirvanalabs.io/faq/): Common integration and account questions.
- [Dashboard](https://dashboard.nirvanalabs.io): Web console. Its code-generation panel emits equivalent Go, TypeScript, curl and Terraform for any action.
- [Service status](https://status.nirvanalabs.io/): Live incident and uptime reporting.

## Products

- [Product overview](https://nirvanalabs.io/product): How compute, storage, networking and orchestration fit together.
- [Compute](https://nirvanalabs.io/product/compute): Virtual machines and instance types.
- [Agent Sandboxes](https://nirvanalabs.io/product/compute/agent-sandboxes): Isolated environments for running agent-generated code.
- [Storage](https://nirvanalabs.io/product/storage): Adaptive Block Storage — IOPS, latency and durability characteristics.
- [Networking](https://nirvanalabs.io/product/networking): VPCs, firewall rules and private connectivity.
- [Orchestration](https://nirvanalabs.io/product/orchestration): Nirvana Kubernetes Service (NKS).
- [Blockchain nodes](https://nirvanalabs.io/blockchains): Supported chains for managed RPC and archive nodes.

## Evidence & pricing

- [Pricing](https://nirvanalabs.io/pricing): Rates and cost calculator.
- [Benchmark results](https://nirvanalabs.io/benchmark-results): Published methodology and raw numbers versus AWS, GCP and other providers.
- [Case studies](https://nirvanalabs.io/case-studies): Production deployments — Aori, Goldsky, Monad, Optimism.
- [Security & compliance](https://nirvanalabs.io/security): Controls and SOC 2 posture.
- [Service level agreement](https://nirvanalabs.io/sla): Quarterly uptime target of 99.9% for managed RPC node APIs and Nirvana Cloud. Service credits of 10% of the affected quarterly spend for uptime between 99.00% and 99.9%, and 20% below 99.00%, capped at 90 days of paid service per quarter and claimable within 30 days. Standard support responds in under 24 business hours. The page also states the uptime calculation and the five exclusions.

## Use cases

- [AI workloads](https://nirvanalabs.io/use-cases/ai): Agent and inference infrastructure patterns.
- [Web3 workloads](https://nirvanalabs.io/use-cases/web3): Validator, RPC and indexer patterns.
- [Database workloads](https://nirvanalabs.io/use-cases/database): OLTP and OLAP deployment patterns.

## Company

- [About](https://nirvanalabs.io/about): Company background.
- [Contact](https://nirvanalabs.io/contact): Sales and support enquiries.
- [Partners](https://nirvanalabs.io/partners): Technology and infrastructure partners.
- [Changelog](https://nirvanalabs.io/changelog): Dated product release history.
- [Blog](https://nirvanalabs.io/blog): Engineering and product writing.
- [Brand kit](https://nirvanalabs.io/brand-kit): Logos and usage guidance.
- [Terms](https://nirvanalabs.io/terms): Terms of use.
- [Privacy](https://nirvanalabs.io/privacy): Privacy policy.

Full agent index: https://nirvanalabs.io/llms.txt