vizu-client 0.1.0

Vizu is a high-performance observability and tracing platform for LLM agents and AI apps, combining semantic search, specialized evals (RAGAS, G-Eval, toxicity), and Git-like versioning of prompts and responses.

Vizu Python SDK

Python SDK for Vizu - A purpose-built agent trace engine for LLM agents.

Installation

pip install vizu

With Framework Integrations

# Install specific framework integrations
pip install vizu[langchain]      # LangChain / LangGraph
pip install vizu[llamaindex]     # LlamaIndex
pip install vizu[openai-agents]  # OpenAI Agents SDK
pip install vizu[autogen]        # Microsoft AutoGen
pip install vizu[semantic-kernel] # Semantic Kernel
pip install vizu[crewai]         # CrewAI
pip install vizu[smolagents]     # Hugging Face smolagents
pip install vizu[pydantic-ai]    # PydanticAI
pip install vizu[strands]        # AWS Strands Agents
pip install vizu[google-adk]     # Google ADK

# Or install all framework integrations at once
pip install vizu[all-frameworks]

# For development (includes all frameworks + dev tools)
pip install vizu[all]

Quick Start

from vizu import VizuClient, SpanType

# Initialize client
client = VizuClient(
    url="http://localhost:8080",
    tenant_id=1,
    project_id=0
)

# Log a trace
with client.trace(span_type=SpanType.ROOT) as root:
    # Planning step
    with root.child(SpanType.PLANNING) as planning:
        planning.set_token_count(50)
        planning.set_confidence(0.95)
    
    # Tool call
    with root.child(SpanType.TOOL_CALL) as tool:
        tool.set_token_count(20)
        tool.set_duration_ms(150)
    
    # Final response
    with root.child(SpanType.RESPONSE) as response:
        response.set_token_count(80)
        response.set_confidence(0.94)

# Query traces
edges = client.query_temporal_range(
    start_timestamp_us=start_time,
    end_timestamp_us=end_time
)

# Get causal relationships
children = client.get_children(edge_id)
ancestors = client.get_ancestors(edge_id)

Framework Integrations

Vizu provides seamless integrations for all major AI agent frameworks:

Supported Frameworks

Framework	Install	Documentation
LangChain / LangGraph	pip install vizu[langchain]	Chains, agents, workflows
LlamaIndex	pip install vizu[llamaindex]	Query engines, agents, workflows
OpenAI Agents SDK	pip install vizu[openai-agents]	Agent wrappers, sessions
Microsoft AutoGen	pip install vizu[autogen]	Multi-agent conversations
Semantic Kernel	pip install vizu[semantic-kernel]	Kernel functions, planners
CrewAI	pip install vizu[crewai]	Crews, tasks, collaboration
Hugging Face smolagents	pip install vizu[smolagents]	Code agents, tool calling
PydanticAI	pip install vizu[pydantic-ai]	Type-safe agents
Strands Agents	pip install vizu[strands]	AWS agents, multi-provider
Google ADK	pip install vizu[google-adk]	Gemini agents

Quick Integration Examples

LangChain

from vizu.integrations.langchain import VizuCallbackHandler
from langchain.chains import LLMChain

callback = VizuCallbackHandler(
    url="http://localhost:8080",
    tenant_id=1
)

chain = LLMChain(llm=llm, callbacks=[callback])
result = chain.run("What is the weather?")

LlamaIndex

from vizu.integrations.llamaindex import create_callback_manager
from llama_index.core import VectorStoreIndex

callback_manager = create_callback_manager(
    vizu_url="http://localhost:8080",
    tenant_id=1
)

index = VectorStoreIndex.from_documents(
    documents,
    callback_manager=callback_manager
)

OpenAI Agents SDK

from vizu.integrations.openai_agents import VizuAgentWrapper
from openai_agents import Agent

agent = Agent(name="assistant", instructions="You are helpful")
wrapped = VizuAgentWrapper(
    agent=agent,
    vizu_url="http://localhost:8080",
    tenant_id=1
)

session = wrapped.create_session()
response = wrapped.run(session, "Hello!")

PydanticAI

from vizu.integrations.pydantic_ai import wrap_pydantic_ai_agent
from pydantic_ai import Agent

agent = Agent("openai:gpt-4")
agent = wrap_pydantic_ai_agent(
    agent,
    vizu_url="http://localhost:8080",
    tenant_id=1
)

result = agent.run_sync("Process this request")

📚 For detailed integration guides and examples, see INTEGRATIONS.md

Features

Core SDK

• Low-level API: Direct control over edge creation and querying
• Context Managers: Pythonic span tracking with automatic parent-child relationships
• Async Support: Full async/await support for high-performance applications
• Type Safety: Full type hints and Pydantic models
• Causal Queries: Navigate agent reasoning graphs
• Semantic Search: Find similar traces using vector embeddings

Framework Integrations (New in v2.0!)

• 10+ Framework Support: LangChain, LlamaIndex, OpenAI Agents, AutoGen, and more
• Automatic Tracking: LLM calls, token usage, costs, and execution timing
• OpenTelemetry GenAI: Full semantic conventions support
• Production Ready: Async/sync, error resilience, connection pooling
• Zero Code Changes: Wrap existing agents with minimal modifications

Documentation

• Quick Start: See above and examples/
• Framework Integrations: INTEGRATIONS.md
• Full API Documentation: https://docs.vizu.dev/python-sdk
• Examples:
  • LangChain/LangGraph
  • LlamaIndex
  • OpenAI Agents
  • All Frameworks Quickstart

License

MIT License