Skip to main content
The full path from building a bot, through publishing a bot provider, to wiring it into a front end with the SDK

Integration guide

This page strings the existing Asgard documentation into one path, from nothing to live: design a bot in Odin, publish it as a bot provider, get the credentials, and connect it to your application with the SDK. Each step links to the detailed page for it, and following them through gets you your first integration.

For front-end developers

If you would rather look at interactive component examples, read this alongside the integration overview on the SDK demo site, which takes the @asgard-js/react view.

1. Understand the concepts, choose an integration pattern​

Decide how the application will connect to Asgard first, because it affects every step after it.

  • Integration overview and how to choose: the terms, the four patterns compared, and how to decide
  • The four patterns: Hosted Embed, Direct Connect, Workflow Auth, Backend Relay

2. Build the bot and the workflow in Odin​

Design the conversation, from the simplest Hello up to an AI chatbot connected to an LLM.

  • Building your first workflow: Hello
  • Building a simple AI chatbot
  • Streaming responses

3. Configure the model​

Let the bot call an LLM.

4. Publish as an integration, and get the endpoint and API key​

Publish the bot you designed as a bot provider. This is the step that gives you the botProviderEndpoint and API key the front-end SDK needs.

  • Creating a generic integration (getting the endpoint and API key)

5. Manage the API key​

Getting, rotating and safely storing a key, and the limits such as rate limiting.

6. Connect it to your application with the SDK​

With the endpoint and credentials in hand, wire up the front end or the back end.

7. Understand the message flow​

To handle streamed responses after integrating, start from the sequence of SSE events.

8. Design how the reply looks​

A bot's reply is carried by a message template and rendered by the front end.

Once you have been through this path, go back to the pattern page for the integration model you chose and read the detail of its auth, history, memory and payload.