praxius-sdk

Python SDK for Praxius — instrument AI agent decisions for governance, compliance, and audit.

pip install praxius-sdk

Quick start

from praxius import Praxius

px = Praxius(
    api_key="prx_your_key",
    endpoint="https://app.praxius.ai",
)

result = px.record_decision(
    event_type="loan_decision",
    outcome="APPROVED",
    confidence=0.94,
    runtime_ms=1200,
    policy_id="LENDING-POLICY-2024",
    gates=[
        {
            "gate_key": "credit_score_check",
            "title": "Credit Score Threshold",
            "passed": True,
            "reason": "Score 742 exceeds minimum of 680",
            "category": "creditworthiness",
        },
    ],
    gaps=[],
    context={"loan_type": "personal", "amount": 25000},
)

print(result.event_id)

Trace context manager

Wrap your agent with automatic timing:

with px.trace("fraud_check") as ctx:
    decision = my_agent.run(transaction)
    ctx.record(
        outcome=decision.label,
        confidence=decision.score,
        gates=decision.gate_results,
        gaps=decision.missing_fields,
    )
# Decision is sent when the with-block exits.
# If an exception is raised, an ERROR event is recorded automatically.

Human review events

px.record_review(
    trace_id=original_decision_trace_id,
    action="override",           # "accept" | "override" | "modify" | "escalate"
    original_outcome="DENIED",
    final_outcome="APPROVED",
    rationale="Client provided supplemental income documentation.",
    reviewer_role="senior_underwriter",
    assigned_at="2026-05-12T14:00:00Z",
    completed_at="2026-05-12T14:22:00Z",
)

Typed gate evaluations

Use the GateEvaluation dataclass for full IDE completion:

from praxius import GateEvaluation

gates = [
    GateEvaluation(
        gate_key="aml_screening",
        title="AML Counterparty Screening",
        passed=False,
        reason="Destination account matched OFAC watchlist",
        category="compliance",
        policy_id="AML-POLICY-2024",
        policy_version="v3",
    ),
]

Async support

pip install praxius-sdk[async]

from praxius import AsyncPraxius

px = AsyncPraxius(api_key="prx_...")

async def handle():
    result = await px.record_decision(
        event_type="fraud_check",
        outcome="FLAGGED",
        confidence=0.87,
        runtime_ms=230,
    )

Configuration

Docs

Full documentation at docs.praxius.ai.