Skip to content

Class ZCL_LLM_CLIENT_OLLAMA

AI Generated documentation.

Overview

Implements Ollama LLM client functionality by extending base LLM client (zcl_llm_client_base). Key features:

  • Public Methods:
  • get_client: Factory method creating preconfigured client instances

  • constructor: Initializes client with configs and performs auth setup

  • Core Capabilities:

  • Handles Ollama-specific API communication
  • Supports structured JSON output generation
  • Implements tool call functionality for LLM responses
  • Manages authentication with optional API keys
  • Processes Ollama-specific response formats

Dependencies

  • zcl_llm_http_client_wrapper: HTTP client management
  • zcl_llm_common: JSON serialization/deserialization utilities
  • zcx_llm_validation/zcx_llm_authorization: Exception classes
  • BAdI ZLLM_BADI: Encryption implementation for credentials
  • /ui2/cl_json: JSON handling for tool call arguments

Details

Authentication Flow

Request Building Logic

Key aspects of build_request_json:

  1. Modifies last message for structured output by appending JSON schema
  2. Adds Ollama-specific parameters:
  3. stream: false for non-streaming responses
  4. format parameter when using structured output
  5. Custom options from request parameters

Response Handling

handle_http_response implements:

  • Error mapping for HTTP status codes
  • Tool call processing with argument deserialization
  • Token usage statistics from Ollama-specific fields
  • Structured output parsing when enabled

Tool Call Processing:

LOOP AT response-message-tool_calls
  IF tool matches registered tools
    CREATE DATA func_result TYPE HANDLE parameters
    DESERIALIZE arguments to func_result
    ADD TO result-choice-tool_calls
  ELSE IF required tool missing
    SET ERROR

Type Mappings

Custom Ollama types ensure proper API communication:

  • ollama_response: Matches Ollama's chat response format
  • ollama_tool_call: Handles nested tool call structure
  • ollama_message: Extends standard messages with tool calls table

Structured Output Handling

Special treatment in both request building and response parsing:

  1. Modifies prompt to include JSON schema
  2. Adds format parameter to request
  3. Uses dedicated parser for JSON responses
  4. Handles potential parsing errors with specific error messages