Skip to content

Quickstart: Getting Started with JikiOrchestrator

Just a few lines of code and you're up and running:

# Import the main factory function
from jiki import Jiki

# 1. Create the orchestrator with default settings
# This implicitly uses the default calculator server if available
jiki_instance = Jiki()

# 2. Ask a simple question synchronously
answer = jiki_instance.process("What is 2 + 3?")
print(answer)  # → 5

# 3. Or request a detailed response with tool call trace
detailed = jiki_instance.process_detailed("Multiply 7 by 6")
print(detailed.result)             # → 42
for call in detailed.tool_calls:
    print(f"Tool: {call.tool_name}, Args: {call.arguments}, Result: {call.result}")

Tip: You can also run python -m jiki.cli process "Your question" from the CLI or build your own interface using jiki_instance.run_ui().

Under the Hood: Core Interfaces

Behind this simple API, JikiOrchestrator (which Jiki() creates and configures) depends only on minimal abstractions:

  • Prompt Builder (IPromptBuilder)
  • MCP Client (IMCPClient)
  • Tool Validation (parse_tool_call_content, validate_tool_call)
  • Utilities (context trimming, token counting, output cleaning)

All streaming, JSON‑tag interception, schema validation, and tool invocation happen behind the scenes so you can focus on your application logic.