sumiki 0.1.0

Intuitive Python SDK for Lyzr Agent API - Build AI agents with ease

Lyzr SDK

The official Python SDK for Lyzr - Build production-ready AI agents in minutes

Create powerful AI agents with a clean, intuitive API. Supports multiple LLM providers, knowledge bases (RAG), structured outputs, and streaming.

✨ Features

• 🚀 Simple & Intuitive - agent.run("message") - that's it!
• 🎯 Type-Safe - Full Pydantic v2 validation with autocomplete
• 🔌 Modular - Use what you need, extend what you want
• 🤖 Multi-Provider - OpenAI, Anthropic, Google, Groq, Perplexity, AWS Bedrock
• 📚 Knowledge Bases - Built-in RAG for documents, PDFs, websites
• 🎨 Structured Outputs - Type-safe responses with Pydantic models
• ⚡ Streaming - Real-time response streaming
• 📦 Production-Ready - Robust error handling, retries, validation

📦 Installation

pip install sumiki

🔑 Getting Your API Key

Get your free API key from studio.lyzr.ai

Set it as an environment variable:

export LYZR_API_KEY="your-api-key-here"

🚀 Quick Start

from lyzr import Studio

# Initialize (uses LYZR_API_KEY from environment)
studio = Studio()

# Or pass API key directly
studio = Studio(api_key="your-api-key-here")

# Create an agent
agent = studio.create_agent(
    name="Customer Support Agent",
    provider="openai/gpt-4o-mini"
)

# Run the agent
response = agent.run("What are your business hours?")
print(response.response)

📖 Core Features

1. Smart Agents

Agents are intelligent objects with built-in methods:

# Create
agent = studio.create_agent(
    name="Sales Assistant",
    provider="openai/gpt-4o",
    temperature=0.7
)

# Run
response = agent.run("Hello!")

# Stream
for chunk in agent.run("Tell me about your products", stream=True):
    print(chunk.content, end="")

# Update
agent = agent.update(temperature=0.5)

# Clone
new_agent = agent.clone("Sales Assistant V2")

# Delete
agent.delete()

2. Structured Outputs with Pydantic

Get type-safe, validated responses:

from pydantic import BaseModel
from typing import Literal

class SentimentAnalysis(BaseModel):
    sentiment: Literal["positive", "negative", "neutral"]
    confidence: float
    reasoning: str

agent = studio.create_agent(
    name="Sentiment Analyzer",
    provider="openai/gpt-4o",
    response_model=SentimentAnalysis
)

# Get typed response
result: SentimentAnalysis = agent.run("I love this product!")
print(result.sentiment)  # "positive" - fully typed!
print(result.confidence)  # 0.95

3. Knowledge Bases (RAG)

Create knowledge bases and use them with agents:

# Create knowledge base
kb = studio.create_knowledge_base(
    name="product_docs",
    vector_store="qdrant"
)

# Add documents
kb.add_pdf("user_manual.pdf")
kb.add_docx("policies.docx")
kb.add_website("https://docs.mycompany.com", max_pages=50)
kb.add_text("FAQ: Business hours are 9am-5pm", source="faq.txt")

# Query directly
results = kb.query("How do I reset my password?", top_k=3)

# Use with agent at runtime
agent = studio.create_agent(
    name="Support Bot",
    provider="openai/gpt-4o"
)

response = agent.run(
    "What are your business hours?",
    knowledge_bases=[kb]  # ← Pass KB at runtime
)

4. Multiple Providers

# OpenAI
agent = studio.create_agent(name="Bot", provider="openai/gpt-4o")

# Anthropic
agent = studio.create_agent(name="Bot", provider="anthropic/claude-sonnet-4-5")

# Google
agent = studio.create_agent(name="Bot", provider="google/gemini-2.5-pro")

# Auto-detect provider
agent = studio.create_agent(name="Bot", provider="gpt-4o")

🌟 Supported Models

LLM Providers

• OpenAI: GPT-4o, GPT-5, o3, o4-mini
• Anthropic: Claude Sonnet 4.5, Claude Opus 4.5
• Google: Gemini 2.5 Pro, Gemini 3 Pro Preview
• Groq: Llama 3.3, Llama 4
• Perplexity: Sonar, Sonar Pro (with web search)
• AWS Bedrock: Amazon Nova, Claude via Bedrock

Vector Stores (Knowledge Bases)

• Qdrant, Weaviate, PG-Vector, Milvus, Amazon Neptune

📚 Examples

See the /examples folder for complete working examples:

• quickstart.py - Basic agent creation and usage
• structured_responses.py - Pydantic models for typed outputs
• knowledge_base_basic.py - Creating and managing knowledge bases
• kb_with_agent.py - Using KBs with agents at runtime

🛠️ Advanced Usage

Multiple Knowledge Bases

products_kb = studio.create_knowledge_base(name="products")
policies_kb = studio.create_knowledge_base(name="policies")

# Use multiple KBs with custom config
response = agent.run(
    "What's the refund policy for Product X?",
    knowledge_bases=[
        products_kb.with_config(top_k=5, score_threshold=0.8),
        policies_kb.with_config(top_k=3, retrieval_type="mmr")
    ]
)

Session Management

# Auto-generated session
response = agent.run("Hello")

# Explicit session for continuity
session_id = "user_123"
response1 = agent.run("Question 1", session_id=session_id)
response2 = agent.run("Follow-up", session_id=session_id)

🐛 Error Handling

from lyzr.exceptions import (
    AuthenticationError,
    ValidationError,
    NotFoundError,
    InvalidResponseError
)

try:
    agent = studio.create_agent(name="Bot", provider="invalid/model")
except ValidationError as e:
    print(f"Error: {e.message}")

🤝 Contributing

Contributions welcome! Please submit a Pull Request.

📄 License

MIT License - see LICENSE file

🔗 Links

• Get API Key: studio.lyzr.ai
• Documentation: docs.lyzr.ai
• GitHub: github.com/pradipta-lyzr/lyzr-sdk
• PyPI: pypi.org/project/sumiki
• Website: lyzr.ai

💬 Support

• Email: [e-mail]
• GitHub Issues: github.com/pradipta-lyzr/lyzr-sdk/issues

---

Built with ❤️ by Lyzr