Block PII, forbidden keywords, and policy violations in real-time. Drop-in SDK for LangChain, CrewAI, AutoGPT, OpenAI, and custom agents. One line to guard every output.
Paste agent output below. Our client-side simulator runs the same PII and keyword checks as the real API.
Results will appear here
Works with your stack
Built for teams that take AI governance seriously
Trace every step in multi-agent workflows. Track CrewAI crews, LangChain chains, AutoGPT agents, and OpenAI calls with distributed trace IDs and parent-child span linking.
Learn more →Intercept agent outputs before they reach users. Block PII, forbidden keywords, and policy violations in milliseconds. Not just log them after the fact.
Learn more →When a violation is detected, the output is stopped in its tracks. Email or webhook notifications sent instantly — no surprises at audit time.
Learn more →Generate audit-ready CSV or JSON reports for any date range. Export logs with full metadata and compliance flags.
Learn more →PostgreSQL-backed with row-level isolation, encrypted at rest, and full audit trails. Ready for compliance frameworks.
Learn more →REST API schema aligned with MCP tool-call payloads. Route agent events to AgentAudit without reformatting — works with any MCP-compatible agent.
Learn more →One line of code. Automatic interception. Block violations before they reach users.
import agentaudit
# Initialize with your API key
audit = agentaudit.AgentAudit(api_key="aa_live_...")
# Guardrail: intercept and block violations before delivery
result = audit.guardrail(
action="prompt_submitted",
prompt="User: My SSN is 123-45-6789",
response="Here is your account info..."
)
if not result.allowed:
raise ValueError("Output blocked: " + ", ".join(result.violations))
# Violations blocked. Clean output delivered to user.import { AgentAudit } from 'agentaudit-client';
// Initialize with your API key
const audit = new AgentAudit({
apiKey: 'aa_live_...'
});
// Guardrail: intercept and block violations before delivery
const result = await audit.guardrail({
action: 'prompt_submitted',
prompt: 'User: My SSN is 123-45-6789',
response: 'Here is your account info...'
});
if (!result.allowed) {
throw new Error('Output blocked: ' + result.violations.join(', '));
}
// Violations blocked. Clean output delivered to user.curl -X POST https://agentaudit-api-production.up.railway.app/api/v1/audit-logs \
-H "X-API-Key: aa_live_..." \
-H "Content-Type: application/json" \
-d '{
"action": "prompt_submitted",
"prompt": "User: My SSN is 123-45-6789",
"response": "Here is your account info...",
"checkType": "realtime"
}'
# Response: {"id": "...", "complianceFlags": ["CRITICAL_pii_detect_SSN"]}
# Your code checks complianceFlags and blocks before delivery.One line of code. No refactoring. Works with whatever framework you already use.
from agentaudit import AgentAuditCallbackHandler
from langchain_openai import ChatOpenAI
# One callback guards every LangChain call
llm = ChatOpenAI(model="gpt-4o", callbacks=[AgentAuditCallbackHandler(
api_key="aa_your_key_here",
agent_id="uuid-of-your-agent",
guard=True
)])
llm.invoke("What is the weather?")
# Automatically logged + guardedfrom crewai import Crew, Agent, Task
from agentaudit_crewai import AgentAuditObserver
# One line to audit entire crew execution
observer = AgentAuditObserver(
api_key="aa_your_key_here",
crew_name="Research Crew",
guard=True # Block violations automatically
)
crew = Crew(
agents=[researcher, writer],
tasks=[task1, task2],
callbacks=[observer]
)
crew.kickoff()
# Every task traced, violations blockedfrom agentaudit import AgentAuditAutoGPT, ComplianceViolationAutoGPT
# One decorator guards your entire agent
@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}")
# Automatically traced + guardedfrom agentaudit import AgentAuditOpenAI
# One wrapper guards every OpenAI call
client = AgentAuditOpenAI(
openai_api_key="sk-...",
api_key="aa_your_key_here",
agent_id="uuid-of-your-agent",
guard=True # Block violations automatically
)
response = client.chat_completions_create(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello!"}]
)
# Every call traced, violations blockedStart free. Scale as your agents grow. No seat-based pricing.
No configuration required. Pick your industry, click activate, done.
Detects patient identifiers, medical record numbers, and HIPAA-sensitive terms.
Activate PackFlags payment data, account numbers, and financial compliance violations.
Activate PackFlags harmful outputs, jailbreak attempts, and model-exploitation patterns in real-time.
Activate PackProtects PII and flags data privacy policy violations automatically.
Activate PackStart guarding every agent output in under 60 seconds. Free for side projects.
No credit card required. Start auditing in 60 seconds.