QUASENTRA DOCSPYTHON SDK · v1

Build agents that can act.
Control what executes.

Integrate runtime authorization into locally built AI agents. Your code declares identity, tools, and intended permissions; Quasentra enforces the exact operation and records the outcome.

Start the quickstart →Understand the security model →
TERMINALpip install quasentra
01
CHOOSE YOUR PATH

Meet your agent where it is.

LC
LangChain

Protect tools used with create_agent.

→LG
LangGraph

Add middleware before graph compilation.

→CR
CrewAI

Wrap local BaseTool execution.

→OA
OpenAI Agents

Secure local FunctionTool lifecycle.

→
02
CORE WORKFLOW

Declare once. Enforce every time.

  1. 01
    Create a scoped key

    Grant only registration, policy sync, and decision scopes required by the service.

  2. 02
    Wrap the agent factory

    Define agent identity and intended permissions beside the tools they govern.

  3. 03
    Start the application

    Discovery and policy synchronization run idempotently at startup.

  4. 04
    Observe real decisions

    Review allowed, denied, and approval-required actions in the dashboard.

secure_agent.pyPYTHON
from quasentra import Quasentra

security = Quasentra(api_key=API_KEY)

@security.protect(
    "support-agent",
    permissions={
        "customer.read": "ALLOW",
        "email.send": "REQUIRE_APPROVAL",
        "customer.delete": "DENY",
    },
)
def build_agent():
    return SupportAgent(tools=tools)
Follow the complete start-to-production path →
SECURITY BOUNDARY

Know what the SDK can—and cannot—control.

Quasentra protects supported tool operations that pass through its enforcement path. Local interception cannot protect a second unwrapped tool or credentials retained directly by a compromised process.

Move high-impact credentials and execution behind the trusted gateway.

Explore gateway enforcement →