trusera-sdk 0.2.0

Python SDK for monitoring AI agents with Trusera

Trusera Python SDK

Python SDK for monitoring AI agents with Trusera. Track LLM invocations, tool calls, data access, and more to ensure your AI agents are secure and compliant.

Installation

pip install trusera-sdk

Optional Dependencies

For framework-specific integrations:

# LangChain integration
pip install trusera-sdk[langchain]

# CrewAI integration
pip install trusera-sdk[crewai]

# AutoGen integration
pip install trusera-sdk[autogen]

# Development tools
pip install trusera-sdk[dev]

Quick Start

from trusera_sdk import TruseraClient, Event, EventType

# Initialize the client
client = TruseraClient(api_key="tsk_your_api_key")

# Register your agent
agent_id = client.register_agent(
    name="my-agent",
    framework="custom",
    metadata={"version": "1.0.0"}
)

# Track events
client.track(Event(
    type=EventType.TOOL_CALL,
    name="web_search",
    payload={"query": "latest AI news"},
    metadata={"duration_ms": 250}
))

# Events are automatically flushed in batches
# Manual flush if needed
client.flush()

# Clean up
client.close()

Using the Decorator

The @monitor decorator automatically tracks function calls:

from trusera_sdk import TruseraClient, monitor, set_default_client, EventType

# Set up client
client = TruseraClient(api_key="tsk_your_api_key")
client.register_agent("my-agent", "custom")
set_default_client(client)

# Decorate your functions
@monitor(event_type=EventType.TOOL_CALL)
def search_database(query: str) -> list[dict]:
    # Your implementation
    return [{"id": 1, "title": "Result"}]

@monitor(event_type=EventType.LLM_INVOKE, name="gpt4_call")
async def call_llm(prompt: str) -> str:
    # Works with async functions too
    return "AI response"

# Function calls are automatically tracked
results = search_database("user query")
response = await call_llm("What is AI?")

Framework Integrations

LangChain

from langchain.llms import OpenAI
from langchain.agents import initialize_agent, Tool
from trusera_sdk import TruseraClient
from trusera_sdk.integrations.langchain import TruseraCallbackHandler

# Initialize Trusera
client = TruseraClient(api_key="tsk_your_api_key")
client.register_agent("langchain-agent", "langchain")
handler = TruseraCallbackHandler(client)

# Use with LangChain
llm = OpenAI(callbacks=[handler])
agent = initialize_agent(
    tools=[...],
    llm=llm,
    callbacks=[handler]
)

# All LLM calls and tool usage are tracked
agent.run("Your query here")

CrewAI

from crewai import Crew, Agent, Task
from trusera_sdk import TruseraClient
from trusera_sdk.integrations.crewai import TruseraCrewCallback

# Initialize Trusera
client = TruseraClient(api_key="tsk_your_api_key")
client.register_agent("crew-agent", "crewai")
callback = TruseraCrewCallback(client)

# Create your crew
researcher = Agent(role="Researcher", goal="Research topics")
task = Task(description="Research AI trends", agent=researcher)

crew = Crew(
    agents=[researcher],
    tasks=[task],
    step_callback=callback.step_callback
)

# Execute with tracking
result = crew.kickoff()

AutoGen

import autogen
from trusera_sdk import TruseraClient
from trusera_sdk.integrations.autogen import TruseraAutoGenHook

# Initialize Trusera
client = TruseraClient(api_key="tsk_your_api_key")
client.register_agent("autogen-agent", "autogen")
hook = TruseraAutoGenHook(client)

# Create AutoGen agents
assistant = autogen.AssistantAgent(
    name="assistant",
    llm_config={"model": "gpt-4"}
)

# Register hook
hook.setup_agent(assistant)

# All interactions are tracked
user_proxy = autogen.UserProxyAgent(name="user")
user_proxy.initiate_chat(assistant, message="Hello")

Event Types

The SDK supports tracking various types of agent activities:

• EventType.TOOL_CALL - Tool or function invocations
• EventType.LLM_INVOKE - LLM API calls
• EventType.DATA_ACCESS - Database queries, file reads
• EventType.API_CALL - External API requests
• EventType.FILE_WRITE - File system modifications
• EventType.DECISION - Agent decision points

Configuration

Client Options

client = TruseraClient(
    api_key="tsk_your_api_key",
    base_url="https://api.trusera.dev",  # Optional, defaults to production
    flush_interval=5.0,                    # Seconds between auto-flushes
    batch_size=100,                        # Events per batch
    timeout=10.0,                          # HTTP request timeout
    max_retries=3                          # Retries before dropping events
)

Context Manager

Use the client as a context manager for automatic cleanup:

with TruseraClient(api_key="tsk_your_api_key") as client:
    client.register_agent("my-agent", "custom")
    # ... track events ...
# Automatically flushed and closed

Async Client

For asyncio applications, use AsyncTruseraClient:

from trusera_sdk import AsyncTruseraClient, Event, EventType

async with AsyncTruseraClient(api_key="tsk_your_api_key") as client:
    await client.register_agent("my-agent", "custom")

    client.track(Event(
        type=EventType.TOOL_CALL,
        name="async_search",
        payload={"query": "test"}
    ))

    await client.flush()

Best Practices

1. Use Context Manager: Ensures events are flushed on exit
2. Set Agent ID Early: Call register_agent() or set_agent_id() before tracking
3. Batch Operations: Let the SDK handle batching automatically
4. Sensitive Data: Use capture_args=False in @monitor for sensitive functions
5. Error Handling: The SDK logs errors but won't crash your application

Environment Variables

# Optional: Set default API key
export TRUSERA_API_KEY=tsk_your_api_key

# Optional: Custom API endpoint
export TRUSERA_API_URL=https://api.trusera.dev

Development

# Clone the repository
git clone https://github.com/Trusera/trusera-agent-sdk.git
cd trusera-agent-sdk

# Install development dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linter
ruff check .

# Type checking
mypy trusera_sdk

Documentation

Full documentation is available at docs.trusera.dev/sdk/python

Support

• Website: trusera.dev
• Documentation: docs.trusera.dev
• Issues: GitHub Issues
• Email: [e-mail]

License

Apache License 2.0 - see LICENSE file for details.

Contributing

Contributions are welcome! Please read our Contributing Guide for details on our code of conduct and the process for submitting pull requests.

---

Built with care by the Trusera team. Making AI agents secure and trustworthy.