AgentAudit Python SDK

Official Python SDK for the AgentAudit API — audit logging and compliance monitoring for AI agents.

Installation

pip install agentaudit-client

With LangChain support:

pip install agentaudit-client[langchain]

Quick Start

from agentaudit import AgentAudit

# Initialize
audit = AgentAudit(api_key="aa_your_key_here")

# Log an agent action
audit.log(
    action="prompt_submitted",
    prompt="What is the weather?",
    response="It is sunny today.",
    metadata={"model": "gpt-4", "tokens": 150}
)

LangChain Integration

pip install agentaudit-client[langchain]

from agentaudit import AgentAuditCallbackHandler
from langchain_openai import ChatOpenAI

audit_handler = AgentAuditCallbackHandler(
    api_key="aa_your_key_here",
    agent_id="uuid-of-your-agent",
    guard=True   # default
)

# Use with any LangChain component
llm = ChatOpenAI(model="gpt-4o", callbacks=[audit_handler])
llm.invoke("What is the weather?")
# Automatically logged + guarded to AgentAudit!

Async chains and agents

Use AgentAuditAsyncCallbackHandler with ainvoke so API calls do not block the event loop:

from agentaudit import AgentAuditAsyncCallbackHandler
from langchain_openai import ChatOpenAI

handler = AgentAuditAsyncCallbackHandler(
    api_key="aa_your_key_here",
    agent_id="uuid-of-your-agent",
    guard=True,
)

llm = ChatOpenAI(model="gpt-4o", callbacks=[handler])
await llm.ainvoke("What is the weather?")
print(handler.trace_id)

OpenAI Integration

pip install agentaudit-client[openai]

from agentaudit import AgentAuditOpenAI

client = AgentAuditOpenAI(
    openai_api_key="sk-...",
    api_key="aa_your_key_here",
    agent_id="uuid-of-your-agent",
    guard=True,  # default
)

response = client.chat_completions_create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What is the weather?"}]
)
# Automatically logged + guarded!
print(client.trace_id)

AutoGPT Integration

from agentaudit import AgentAuditAutoGPT

logger = AgentAuditAutoGPT(
    api_key="aa_your_key_here",
    agent_name="MyAutoGPT",
    guard=True,
)

with logger.trace() as trace:
    result = agent.run("Research topic X")
    trace.log_action("run", prompt="Research topic X", response=result)

print(trace.trace_id)

Decorator

from agentaudit import AgentAuditAutoGPT, ComplianceViolationAutoGPT

@AgentAuditAutoGPT.guard(api_key="aa_your_key_here", agent_name="MyAutoGPT")
def run_agent(task: str) -> str:
    return agent.run(task)

try:
    result = run_agent("Research topic X")
except ComplianceViolationAutoGPT as e:
    print(f"Blocked: {e.violations}")

Features

• Simple logging: One-line audit log submission
• Automatic compliance: PII detection, keyword matching, rate limiting, regex matching, sentiment analysis, custom validators
• Agent registration: Track which agents are performing actions
• Query and export: Retrieve audit logs with filters
• LangChain support: Drop-in callback handler
• Type hints: Full type annotation support

Agent-to-Agent Audit Trails

Track multi-agent conversations and CrewAI workflows with distributed tracing:

import uuid
from agentaudit import AgentAudit

audit = AgentAudit(api_key="aa_your_key_here")

# Start a trace — e.g. when a CrewAI crew begins execution
trace_id = str(uuid.uuid4())

# Log the root event (crew start)
root = audit.log(
    action="crewai_crew_start",
    trace_id=trace_id,
    metadata={"crew": "Research Crew", "task_count": 3}
)

# Log child events (tasks, agent actions) with parent_span_id
audit.log(
    action="crewai_task_start",
    trace_id=trace_id,
    parent_span_id=root.id,
    prompt="Research topic X",
    metadata={"task_id": "task-1"}
)

# Query the full trace later
# (use the HTTP client or dashboard to query by traceId)

CrewAI Integration

The CrewAI observer (agentaudit.crewai) automatically manages trace IDs and parent span IDs:

from agentaudit import AgentAuditObserver

observer = AgentAuditObserver(api_key="aa_key", crew_name="My Crew")
# trace_id is generated automatically in on_crew_start
# every event shares the same trace_id with proper parent_span_id linking

Documentation

Full API documentation: https://agentaudit.online/docs

License

MIT

Publishing (Maintainers Only)

This package is published automatically via GitHub Actions when you push a version tag:

Prerequisites

1. Create a PyPI account
2. Generate an API token at pypi.org/manage/account/token
3. Add the token to your GitHub repo: Settings → Secrets and variables → Actions → New repository secret
   • Name: PYPI_API_TOKEN
   • Value: your PyPI API token (starts with pypi-)

Publish a New Version

# Update version in sdk/python/setup.py
git add sdk/python/setup.py
git commit -m "chore: bump Python SDK to v1.0.1"
git tag v1.0.1
git push origin v1.0.1

The publish-python.yml workflow will automatically build and upload to PyPI.