Skip to main content
For the complete documentation index for agents and LLMs, see llms.txt.

Deployment Options

You can run Haystack Enterprise Platform as a managed cloud service or on your own infrastructure, including on-premises and air-gapped environments. The option you choose determines where your files, pipelines, traces, and models run, and what leaves your network.

Haystack Enterprise Platform is the platform for sovereign AI: you decide where your AI applications run, who operates the infrastructure, and which models they use.

Deployment Options at a Glance​

This table compares the options by where the platform runs, where your data is stored, and which models you can use:

OptionWhere the platform runsWhere your data is storedModels you can use
Managed cloudInfrastructure operated by deepsetStorage managed by deepsetAny supported model provider
Self-hostedYour private cloud or on-premises data centerYour infrastructureModel providers you allow and models you host yourself
Air-gappedYour infrastructure, with no connection to the internetYour infrastructureModels you host inside your network

All options use the same platform: you build pipelines and agents in Builder, deploy them, and call them through the REST API in the same way. To learn how to deploy a pipeline, see Deploy a Pipeline.

Managed Cloud​

In the managed cloud, deepset runs and maintains the platform. You sign in at the URL for your chosen cloud provider and region, and the platform handles deployment, scaling, and updates. Pipelines scale automatically with traffic.

Files you upload and the document stores the platform manages are stored in deepset's infrastructure. To learn how data moves through the platform, see Data Flow in Haystack Enterprise Platform.

Self-Hosted and On-Premises​

You can run the platform in your own environment: a private cloud account or an on-premises data center. In a self-hosted deployment, the platform, your files, document stores, traces, and logs all stay in infrastructure you control. You decide which external services, if any, the platform can reach.

deepset works with infrastructure partners for self-hosted deployments, including AWS, StackIT, NVIDIA, and on-premises data center providers.

Air-Gapped Deployments​

An air-gapped deployment is a self-hosted deployment with no connection to the internet. Nothing leaves your network: not your documents, queries, prompts, traces, or model inputs and outputs.

Because the platform can't reach external services, plan for these differences:

  • Models: Hosted model APIs, such as OpenAI or Anthropic, aren't available. deepset doesn't run models for you. Host your LLMs, embedding models, and ranking models inside your network separately, then connect them to the platform. Custom model definitions only describe the connection to a model you host, not the model itself. For details, see Add Custom Model Definitions.
  • External services: Components that call third-party APIs, such as web search, translation, or hosted file conversion services, work only if you host an equivalent service inside your network.
  • Observability: Traces and logs are stored inside the air-gapped environment, so you can access them only from within that environment. You can also use an observability tool you host yourself.

Data Residency and Sovereignty​

Data residency requirements specify where data is stored and processed, for example, only within the European Union (EU). Sovereignty requirements go further and specify who operates the infrastructure and which jurisdictions it falls under.

Haystack Enterprise Platform is built for sovereign AI. You keep control over where your data is processed, who operates the platform, and which models your pipelines and agents call.

The deployment option you choose determines how you meet these requirements:

  • With a self-hosted or air-gapped deployment, you choose the location and operator of every part of the system, including the models.
  • In every option, you control which model providers your pipelines call. Data sent to an external model provider is processed under that provider's terms. For details, see Data Privacy and LLM Providers.

Choosing an Option​

Consider these questions when you choose a deployment option:

  • Can your documents leave your network? If not, use a self-hosted or air-gapped deployment.
  • Can your prompts and documents be sent to external model providers? If not, host your models yourself, or use an air-gapped deployment.
  • Do you want deepset to operate and update the platform? If so, use the managed cloud.

To discuss a self-hosted or air-gapped deployment, contact your deepset representative.