> ## Documentation Index
> Fetch the complete documentation index at: https://docs.livepeer.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Gateways Vs. Orchestrators: What’s the Difference?

> Livepeer uses two core node types-**Gateways** and **Orchestrators**-that work together to provide real-time AI video compute at scale. Although closely connected, they serve entirely different purposes. This page breaks down how they differ and why both roles matter for a decentralised compute marketplace.

<CustomDivider />

## Overview

In brief:

<Callout>
  <FlexContainer direction="column" gap="16px" align="center">
    <Icon icon="torii-gate" /> **Gateways coordinate.**

    <Icon icon="microchip" /> **Orchestrators compute.**
  </FlexContainer>
</Callout>

Together, they form the backbone of the Livepeer AI video pipeline.

```mermaid theme={"theme":{"light":"github-light","dark":"dark-plus"}}
sequenceDiagram
    participant App as Application
    participant GW as Gateway
    participant ORC as Orchestrator

    App->>GW: Submit job (model, parameters, video frames)
    GW->>ORC: Route job to best GPU
    ORC->>GW: Return processed frames/output
    GW->>App: Send results back to user
```

| Role             | Function                                       | Performs GPU Work? | External-Facing? |
| ---------------- | ---------------------------------------------- | ------------------ | ---------------- |
| **Gateway**      | Job intake, pricing, routing, capability match | ❌ No               | ✅ Yes            |
| **Orchestrator** | GPU compute, inference, transcoding, BYOC      | ✅ Yes              | ❌ No             |

## Gateway Responsibilities

Gateways act as the front door to the network:

* Receive jobs from applications
* Determine required model, pipeline, or GPU
* Select the best Orchestrator based on performance and pricing
* Route the workload with low latency
* Return results to the client
* Publish marketplace offerings (models, pipelines, cost per frame, etc.)

Gateways provide *service intelligence*, not compute.

<CustomDivider />

## Orchestrator Responsibilities

Orchestrators are GPU operators who run:

* Real-time AI inference
* Daydream / ComfyStream pipelines
* BYOC containers
* Traditional transcoding

They provide:

* GPU horsepower
* Model execution
* Deterministic and verifiable output
* Performance guarantees

They do not expose external APIs directly-Gateways handle that.