SIARE Documentation

Self-Improving Agentic RAG Engine - Evolve multi-agent RAG pipelines using Quality-Diversity optimization.

SIARE treats your RAG pipeline configuration as a searchable space and uses AI-driven evolution to discover optimal strategies for your specific domain and tasks.

---

Quick Links

Getting Started	Core Concepts	Guides
Installation	System Architecture	First Custom Pipeline
Quickstart	Data Models	Prompt Engineering
Configuration	Glossary	Custom Extensions

---

What is SIARE?

Traditional RAG systems require manual tuning of retrieval strategies, prompt templates, and agent configurations. SIARE automates this through evolutionary optimization:

1. Define your multi-agent pipeline (roles, tools, graph structure)
2. Provide evaluation tasks and metrics
3. Evolve - SIARE mutates and evaluates configurations to find what works best
4. Deploy the optimal pipeline for your domain

Key Features

• Quality-Diversity Optimization - MAP-Elites algorithm maintains diverse high-performing solutions
• 6 Mutation Types - Prompt changes, parameter tweaks, topology rewiring, crossover
• Multi-Agent DAGs - Conditional execution paths with role-based specialization
• Extensible Adapters - Vector search, web search, custom tools
• Hook System - Observe and extend core behavior without modifying source

---

Installation

pip install siare

For full features including LLM providers and embeddings:

pip install siare[full]

See Quickstart for detailed setup instructions.

---

Example

from siare import pipeline, role, edge, task

# Define a simple RAG pipeline
config = pipeline(
    name="my-rag",
    roles=[
        role("retriever", "Find relevant documents", model="gpt-4o-mini"),
        role("synthesizer", "Generate answer from context", model="gpt-4o"),
    ],
    edges=[
        edge("retriever", "synthesizer"),
    ],
)

# Define evaluation tasks
tasks = [
    task("What is machine learning?", expected="supervised and unsupervised..."),
]

# Evolve the pipeline
from siare.services import DirectorService, ExecutionEngine

director = DirectorService(llm_provider)
engine = ExecutionEngine(llm_provider)

# Run evolution loop...

See Use Cases for complete examples.

---

Documentation Sections

Getting Started

• Quickstart Guide - Installation and first pipeline
• Configuration - All configuration options
• Troubleshooting - Common issues and solutions

Architecture

• System Architecture - How SIARE works
• Data Models - Core data structures

Guides

• First Custom Pipeline - Build your first pipeline
• Prompt Engineering - Optimize prompts
• Custom Extensions - Extend SIARE
• Use Cases - Real-world examples

Reference

• Glossary - Key terms and concepts
• Why SIARE? - Comparison with alternatives
• Contributing - How to contribute

---

License

SIARE is open source under the MIT License.