AI agents make server-to-server HTTP requests to your backend system. The requests are initiated when a user path triggers an integration node in their dialogue, and the mapped returned data is stored and available for the duration of the AI agents session.
Technical Requirements of your backend API
API Types
Your API must be one of the following architectures:
- REST
- GraphQL
Response
Your API response must follow the general requirements:
- Provide an API response within 30 seconds
- Preferably return JSON format
Authentication
From the AI agents side, we can authenticate using any of the following:
- API key
- Bearer token
- Expiration token
- Basic auth
- OAuth 2.0
- IP safelists / allowlist
Self Checklist
The self-checklist below contains questions and information that will help you build a robust API:
- Is the API "searchable" with data that a typical user would know, or with data passed through from your CRM?
- Does the data in the API response contain information that satisfies your use case?
- If your use case requires data from multiple systems, can you consolidate them into a middle layer?
- Is it known when the API throws a 200, a 400, and a 404 HTTP status code?
- What does the API return if the input is invalid or incorrect?
- Does your system set an API rate limit?
- How long does your system take to send a response?
Please be ready to provide:
• Documentation: Auth, API request query / body and response schema
• (Optionally a Postman collection.)
• Test data for the most common outcomes.
Read more about the onboarding plan in the previous article, Custom integration - preparation.