SDK · Python · TypeScript

When you'd rather enforce inside your own agent loop, the SANSAR SDK wraps each tool call in code.

You get the same checkpoint as the gateway — policy, identity, signed stamp — without a separate network hop. Wrap the tools your agent can reach, and every invocation returns a verdict plus a verifiable evidence record.

Install

Python

# Python 3.9+
pip install sansar

TypeScript / Node

// Node 18+
npm install @sansar/sdk

Wrap a tool call

Python

from sansar import Gate

gate = Gate(agent="billing-bot", key=SANSAR_KEY)

# check before you act
verdict = gate.check(action="refund", args={"amount": 240, "acct": "A-19"})

if verdict.allowed:
    result = tools.refund(amount=240, acct="A-19")
    gate.stamp(verdict, result)   # signed evidence record
else:
    escalate(verdict.reason)      # blocked or sent to a human

TypeScript

import { Gate } from "@sansar/sdk";

const gate = new Gate({ agent: "billing-bot", key: SANSAR_KEY });

const verdict = await gate.check({ action: "refund", args: { amount: 240, acct: "A-19" } });

if (verdict.allowed) {
  const result = await tools.refund({ amount: 240, acct: "A-19" });
  await gate.stamp(verdict, result);   // signed evidence record
} else {
  escalate(verdict.reason);
}

What you get back

• An allow / block / escalate verdict, decided by your policy — before the action runs.
• A signed decision record (Ed25519, HSM-backed) chained into a tamper-evident log.
• The same record any auditor can verify independently, with open tooling and no SANSAR server.