Guides

The platform works in three layers: you upload content as assets, build a knowledge store from that content, and create responses through natural language.

┌─────────────────────────────────┐
│         Your Application        │  ← You build this
├─────────────────────────────────┤
│     Responses API               │  ← Natural language queries
├─────────────────────────────────┤
│     Knowledge Store             │  ← Video + derived understanding
├─────────────────────────────────┤
│     Assets                      │  ← Video, audio, image files
└─────────────────────────────────┘

Set up

Upload content, build a knowledge store, and add assets to it before you create a response.

Upload content

Upload videos to the platform. Choose between direct and multipart uploads.

Create a knowledge store

Create a knowledge store and optionally configure what the platform extracts.

Add assets

Add assets to a knowledge store for indexing.

Create a response

Before you create a response, prepare these resources:

  1. An asset with ready status.
  2. A knowledge store. Ingestion configuration is optional. Add it when you create the knowledge store if you want Jockey to emphasize specific fields, entities, or domain signals during indexing.
  3. A knowledge store item with ready status.

Use the POST /responses endpoint to create a response. The response is shaped by what you ask, the instructions you provide, and the output schema you define. Start with Create a response to learn the base request and response format.

Create a response

Learn the base POST /responses request, how to use instructions, and how to read the output.

After you create a response, use these optional guides to change how the response is delivered, formatted, or continued across turns.

Streaming

Stream response tokens in real time with Server-Sent Events.

Structured output

Return typed JSON from a response by providing a schema.

Multi-turn sessions

Continue response generation across turns by passing a session_id.