Jiki: LLM Tool Orchestration Framework
Jiki connects large language models to tool servers using the Model Context Protocol (MCP), enabling powerful AI applications with just a few lines of code.
What Can You Do With Jiki?
- Build AI Assistants that use specialized tools to solve complex tasks
- Connect to Any LLM Provider through LiteLLM (OpenAI, Anthropic, Google, etc.)
- Create Custom Tools that Jiki can discover and invoke automatically
- Trace and Debug AI interactions with comprehensive logging
Quick Start
from jiki import Jiki
# Create an orchestrator with calculator tools
orchestrator = Jiki(
auto_discover_tools=True,
mcp_script_path="servers/calculator_server.py"
)
# Process a query
result = orchestrator.process("What is 15 * 8?")
print(result) # Output: "15 * 8 = 120"
Documentation
Getting Started
- 🚀 Getting Started Guide - Step-by-step installation and first run
- 💻 CLI Reference - Command-line interface documentation
- 📝 Code Examples - Ready-to-run example scripts
Concepts & Architecture
- 🏛️ Architecture Overview - How Jiki works
- 🧩 Core Interfaces - Main protocols and extension points
- 🔧 Orchestrator - The main engine
- 🤝 MCP Client - Tool discovery and invocation
- 📊 Logging & Tracing - Recording interactions
Reference
- 📚 API Reference - Detailed API documentation
Key Features
| Feature | Description |
|---|---|
| Tool Discovery | Automatically detect available tools from an MCP server |
| Multiple Transport Options | Connect via stdio (subprocess) or HTTP (SSE) |
| Detailed Responses | Get structured data about tool calls and execution traces |
| Conversation State | Save and resume conversations between sessions |
| LLM Flexibility | Works with all major LLM providers via LiteLLM |
Next Steps
After getting familiar with the basics:
- Follow the Getting Started Guide to install and run your first example
- Try the CLI commands for quick interactive sessions
- Explore the code examples for more advanced use cases