Govern every action an AI agent takes on your behalf.
The Agent Governance SDK resolves the authority behind every agent action — admissibility before execution, a signed Evidence Pack™ after. Your agent proposes; the governance engine decides; the SDK proves it. TypeScript · Python · Go.
Authority, not just observability
Adjacent tools watch what a model said. This SDK decides whether an action is allowed to happen — and makes the decision replay-provable from public keys alone.
Admissibility before commit
Every proposed action is checked against purpose, scope, and delegated authority before it is permitted to execute. Fail-closed by default.
Signed Evidence Pack™
Each decision emits a signed, tamper-evident evidence record — replay-derivable from published keys, verifiable offline by an auditor.
Self-governing agents
Every scaffolded agent emits the governance event family by default. Authority in, evidence out — no bolt-on audit step.
Install
Three languages, one wire contract. The verification surface is 100% open.
# TypeScript / JavaScript npm install @kye/agent-dev-kit # Python pip install kye-agent-dev-kit # Scaffold a governed agent npx kye agent new my-agent # emits the §0.3 governance events by default
import { GovernanceClient } from "@kye/agent-dev-kit"; const kye = new GovernanceClient({ trustDomain: "acme" }); // Ask BEFORE the agent acts const verdict = await kye.admissible({ action: "payment.send", purpose: "supplier-invoice", scope: { amount: 4999, currency: "GBP" }, }); if (verdict.admitted) await executePayment(); // verdict.evidencePack is signed + replay-provable
Where the SDK sits in the lifecycle
The Agent Governance SDK covers the authority-resolution and evidence stages of the KYE Agentic Governance Lifecycle™.