vantageai

Python SDK for Vantage governance, execution-boundary enforcement, and brokered model execution.

Install

pip install vantageai

Core Client

from vantageai import VantageClient

client = VantageClient(
    api_key="vt-...",
    agent_id="mortgage-underwriter",
    base_url="https://api.vantage.ai",
)

Evaluate, Authorize, Commit

decision = await client.evaluate(
    action_type="approve_large_loan",
    action_payload={"amount": 850000, "applicantId": "app-123"},
)

authorization = await client.authorize(
    action_type="openai.chat.completions.create",
    action_payload={
        "model": "gpt-4.1-mini",
        "messages": [{"role": "user", "content": "Review this file"}],
    },
    execution_trace_id="trace-123",
    ttl_seconds=120,
)

if authorization.decision == "allow" and authorization.authorization:
    await client.commit_authorization(
        authorization.authorization.id,
        action_payload={
            "model": "gpt-4.1-mini",
            "messages": [{"role": "user", "content": "Review this file"}],
        },
        execution_trace_id="trace-123",
    )

Brokered Provider Execution

result = await client.broker_openai_chat_completion(
    request={
        "model": "gpt-4.1-mini",
        "messages": [{"role": "user", "content": "Review this mortgage application"}],
        "stream": False,
    },
    execution_trace_id="trace-123",
    ttl_seconds=180,
)

if result.decision == "allow":
    print(result.authorizationId, result.commitAttemptId, result.completion)

Anthropic and OpenAI-compatible broker client helpers are available too:

from vantageai import create_brokered_anthropic_client, create_brokered_openai_client

openai = create_brokered_openai_client(client, execution_trace_id="trace-123", ttl_seconds=180)
anthropic = create_brokered_anthropic_client(client, execution_trace_id="trace-123", ttl_seconds=180)

Observed Wrappers

from openai import OpenAI
from vantageai import VantageClient, wrap_openai_client

vantage = VantageClient(api_key="vt-...", agent_id="mortgage-underwriter")
openai = wrap_openai_client(OpenAI(api_key="..."), vantage)

await openai.chat.completions.create(
    model="gpt-4.1-mini",
    messages=[{"role": "user", "content": "Summarize this file"}],
)

Observed wrappers keep post-execution logging behavior. If the org is in broker-required mode, strict logging surfaces that violation instead of silently claiming execution was governed.

Operator Visibility

boundary = await client.get_execution_boundary_settings()
authorizations = await client.list_execution_authorizations(limit=20)
attempts = await client.list_execution_commit_attempts(limit=20)
performance = await client.get_performance_snapshot()

Sync Usage

from vantageai import VantageClientSync

client = VantageClientSync(api_key="vt-...", agent_id="mortgage-underwriter")
snapshot = client.get_execution_boundary_settings()