> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt.
> For full documentation content in one file, see https://docs.sarvam.ai/llms-full.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Managed vs Self-Hosted

> Same Sarvam models, two delivery models. Compare the Managed API at api.sarvam.ai with a self-hosted SageMaker deployment in your own AWS account, and pick the right one.

Sarvam offers the same models two ways. Most teams start on the **Managed API** and move some or all workloads to **self-hosted** when data residency, scale, or cost make it worthwhile.

## At a glance

|                          | Managed (`api.sarvam.ai`)         | Self-Hosted (SageMaker)                                             |
| ------------------------ | --------------------------------- | ------------------------------------------------------------------- |
| **Where it runs**        | Sarvam's cloud                    | Your AWS account and VPC                                            |
| **Where your data goes** | Sent to Sarvam over HTTPS         | Never leaves your cloud                                             |
| **Pricing**              | Per request / per unit            | AWS instance-hours (software fee + infra)                           |
| **Setup**                | API key, one request              | Subscribe, deploy an endpoint                                       |
| **Scaling**              | Managed for you                   | You size and autoscale the endpoint                                 |
| **Best for**             | Getting started, spiky/low volume | Data-residency needs, steady high volume, air-gapped or in-VPC apps |

## Choose the Managed API if…

#### You want to ship today

Grab an API key and make your first call in minutes — no infrastructure to manage.

#### Your volume is low or spiky

Pay only for what you use, with no idle instance cost between bursts.

Start here: [Sarvam API Quickstart](/api/getting-started/quickstart).

## Choose self-hosted if…

#### Data cannot leave your cloud

Compliance, contractual, or regulatory requirements (BFSI, healthcare, government) mean audio and documents must stay in your VPC.

#### You run steady, high volume

At sustained throughput, dedicated instance-hours are more predictable — and often cheaper — than per-call pricing.

#### You need in-network latency

Keep inference next to your application, with no public-internet hop.

#### You deploy into restricted networks

Endpoints run with network isolation — no outbound internet from the model container.

Start here: [Get started on SageMaker](/api/self-hosted/sagemaker/get-started).

> **Note**
>
> You don't have to pick one. A common pattern is the Managed API for development and low-volume traffic, and self-hosted endpoints for production workloads that carry sensitive data or high steady volume.

> Same Sarvam models, two delivery models. Compare the Managed API at api.sarvam.ai with a self-hosted SageMaker deployment in your own AWS account, and pick the right one.