Skip to content

Class ZCL_LLM_CHAT_REQUEST

AI Generated documentation.

Overview

The zcl_llm_chat_request class implements LLM chat request management through interface zif_llm_chat_request, providing structured message/tool handling and configuration for AI interactions.

Core Functionality:

  • Message management: Add/clear individual or bulk messages
  • Tool integration: Register tools and handle tool execution results
  • Request configuration: Control response formatting and tool selection behavior

Public Methods:

Category Methods
Message Handling add_message, add_messages, clear_messages, add_tool_choice
Tool Management add_tool, add_tools, add_tool_result, clear_tools
Configuration set_tool_choice, set_structured_output_active
Data Access get_internal_request, get_messages, get_tools, options

Dependencies

  • Interface: zif_llm_client (role constants)
  • Utility Class: zcl_llm_common (JSON conversion)
  • Data Structure: zllm_request (core container type)
  • Interface: zif_llm_chat_request (implemented contract)

Details

sequenceDiagram
    participant Client
    participant zcl_llm_chat_request
    participant zllm_request
    participant Tool

    Client->>zcl_llm_chat_request: add_message()
    zcl_llm_chat_request->>zllm_request: Append to messages
    Client->>zcl_llm_chat_request: add_tool(Tool)
    zcl_llm_chat_request->>zllm_request: Store tool + set tool_choice
    Client->>zcl_llm_chat_request: add_tool_result(Tool)
    zcl_llm_chat_request->>Tool: get_result()
    Tool-->>zcl_llm_chat_request: tool_result
    zcl_llm_chat_request->>zcl_llm_common: to_json()
    zcl_llm_chat_request->>zllm_request: Add tool message

Key Implementation Aspects:

  1. State Preservation: Maintains complete request context in zllm_request structure including:
  2. Message history with role metadata
  3. Registered tools with automatic tool_choice updates

  4. Structured output configuration flags

  5. Tool Result Handling:

METHOD add_tool_result.
  DATA(tool_result) = tool->get_result().  // Dynamic interface binding
  APPEND VALUE #(role = zif_llm_client=>role_tool,
                content = zcl_llm_common=>to_json(...))  // Structured JSON conversion
ENDMETHOD.
  1. Validation Patterns:
  2. Automatic role assignment for tool responses (role_tool)
  3. Tool call ID propagation through message chain

  4. Immutable request export via get_internal_request

  5. Configuration Cascade:

set_tool_choice(AUTO) -> Affects subsequent tool registrations
set_structured_output_active -> Forces JSON schema validation
add_tool_choice -> Creates assistant role message with tool calls