AI Agents API User Guide

Overview

The AI Agent API lets developers interact with AI agents built and hosted on the MariaDB Cloud platform. This API is perfect for embedding conversational AI experiences in your applications, dashboards, or internal tools – all without the need for complex LLM infrastructure or MLOps.

Authentication

To make requests to the API, you must authenticate using an API key.

Steps:

1. Get an API Key

   • Go to: MariaDB Cloud API Key Management

   • Generate a new API key.

2. Include the key in your request headers:

   X-API-Key: YOUR_API_KEY

Maintaining Chat Sessions

The AI Agent Session API allows you to initiate and manage conversational sessions with your AI agents. Sessions are crucial for maintaining context across multiple interactions, which leads to more natural and coherent conversations with the agent.

Creating a Session

• Endpoint:

  POST https://api.skysql.com/copilot/v1/session

• Headers:

  Content-Type: application/json
  X-API-Key: YOUR_API_KEY

• Request Body:

  {
    "agent_id": "your-agent-id",
    "metadata": {
      "user_id": "optional-user-id",
      "conversation_topic": "optional-topic"
    }
  }

  • agent_id (string): The identifier of the AI agent you want to talk to.

  • metadata (object, optional): Any additional information to associate with the session, like user identifiers or conversation topics.

Example using curl:

curl -X POST https://api.skysql.com/copilot/v1/session \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
        "agent_id": "agent-12345",
        "metadata": {
          "user_id": "user-67890",
          "conversation_topic": "monthly sales report"
        }
      }'

Response Structure

A successful response will return a JSON object containing:

• session_id (string): A unique identifier for the session that was created.

• agent_id (string): The identifier of the agent associated with this session.

• created_at (string): A timestamp indicating when the session was created.

Example Response:

{
  "session_id": "session-abcde12345",
  "agent_id": "agent-12345",
  "created_at": "2025-05-21T18:37:00Z"
}

Using the Session

Once you have a session_id, include it in all subsequent interactions with the agent to maintain context:

• Chat Endpoint:

  POST https://api.skysql.com/copilot/v1/chat

• Request Body:

  {
    "prompt": "Your question here",
    "session_id": "session-abcde12345"
  }

Maintaining the same session_id across multiple requests lets the agent remember previous interactions, leading to more coherent conversations.

Error Handling

If your request is invalid or unauthorized, you might receive error responses such as:

• 401 Unauthorized: Missing or invalid API key.

• 400 Bad Request: Malformed request syntax or missing required fields.

• 500 Internal Server Error: An error occurred on the server.

Always ensure your API key is valid and your request body is correctly formatted.

Tips for Effective Use of Sessions

• Session Management: Use the same session_id for a series of related interactions to keep the conversation flowing.

• Metadata Usage: Leverage the metadata field to store relevant information that can help the agent provide more personalized responses.

• Session Termination: Implement logic in your application to end sessions when they're no longer needed, which helps free up resources.

Making a Chat Request

The Chat API allows you to send natural language prompts to a AI agent and receive structured responses, including generated SQL and data results when applicable. It supports both stateless and multi-turn conversational use cases and lets you pass agent-specific configuration such as table filters to control the context of the response.

The optional config object allows you to customize how the agent accesses and filters data during query generation. This is especially useful for enforcing data policies, scoping context, or running controlled experiments.

IMPORTANT: You can use this API on its own for single interactions, or in conjunction with the Session API to maintain context across multi-turn conversations with the same agent.

• Endpoint:

  POST https://api.skysql.com/copilot/v1/chat

• Headers:

  Content-Type: application/json
  X-API-Key: YOUR_API_KEY

• Request Body:

  {
    "prompt": "Your natural language question",
    "agent_id": "your agent id",
    "session_id": "optional-session-id",
    "config": {
      "table_filter": {
        "table":"column=value"
      }
    }
  }

  • prompt (string): Your question to the AI agent.

  • agent_id (string): The UUID of the AI agent to query

  • session_id (string, optional): Use this to maintain conversational context across turns.

  • config (object, optional): Configuration to pass contextual filters (e.g., row-level security, scope)

  • table_filters (object,optional): Key-value pairs where the key is a table name, and the value is a SQL WHERE clause to apply automatically

Example using curl:

curl -X POST https://api.skysql.com/copilot/v1/chat \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
        "prompt": "What was the total revenue last quarter for the enterprise division?",
        "session_id": "session-4567",
        "agent_id": "your-agent-id",
        "config": {
          "table_filters": {
            "sales": "division = '\''enterprise'\'' AND order_date >= '\''2024-10-01'\'' AND order_date <= '\''2024-12-31'\''"
          }
        }
      }'

Response Format

A successful response returns a structured JSON object:

• response: The agent's natural language reply.

• sql: (if applicable) The SQL generated by the agent to answer the question.

• data: Structured data returned (if relevant).

• session_id: Echoes your session ID, to help manage stateful interactions.

Example Response:

{
  "response": "Here are the top 5 products by revenue last month.",
  "sql": "SELECT product_name, SUM(revenue) FROM sales WHERE date >= '2024-04-01' AND date <= '2024-04-30' GROUP BY product_name ORDER BY SUM(revenue) DESC LIMIT 5;",
  "data": [
    { "product_name": "Widget A", "revenue": 124000 },
    { "product_name": "Widget B", "revenue": 115000 }
  ],
  "session_id": "session-4567"
}

Error Handling

Code	Meaning	Explanation
401	Unauthorized	Missing or invalid API key
400	Bad Request	Malformed request syntax
500	Internal Server Error	Something went wrong on the backend

Best Practices

• Use session_id to maintain conversation context across multiple prompts.

• Check the sql output if you’re debugging or want insight into how the agent is reasoning.

• Use clear, specific language in your prompts for optimal results.

• Validate data before injecting it into your UI or downstream services.

See Also

• API Reference: AI Agent API Docs

• MariaDB Cloud Console: MariaDB Cloud Portal