@lelu-auth/lelu
v0.2.0
Published
TypeScript SDK for Lelu — the authorization engine for autonomous AI agents
Maintainers
Readme
Lelu
The TypeScript SDK for Lelu — the confidence-aware authorization engine for autonomous AI agents.
Lelu provides confidence-aware access control, human-in-the-loop approvals, and SOC 2-ready audit trails for your autonomous agents.
Installation
npm install @lelu-auth/leluStart Dashboard (Localhost)
After installing the package, start the local dashboard stack with one command:
npx @lelu-auth/lelu dashboardThen open:
http://localhost:3002/auditThis command clones/updates the Lelu stack in ~/.lelu-stack and runs docker compose up -d --build.
Or run the public Lelu engine image:
docker pull abenezer0923/lelu-engine:latest
docker run --rm -p 8083:8080 abenezer0923/lelu-engine:latestDocker Support
Lelu works in Dockerized apps.
The SDK resolves engine URL in this order:
baseUrlpassed tocreateClient(...)LELU_BASE_URLenvironment variablehttp://localhost:8080fallback
Example for containers:
LELU_BASE_URL=http://host.docker.internal:8083Quick Start
import { createClient } from "@lelu-auth/lelu";
// Initialize the client
const lelu = createClient({
baseUrl: "http://localhost:8083"
});
// Authorize an agent action
async function runAgent() {
const { allowed, reason } = await lelu.agentAuthorize({
agentId: "agent-123",
action: "read_database",
resource: "users_table",
context: {
confidence: 0.95
}
});
if (allowed) {
console.log("Action permitted!");
} else {
console.log("Action denied:", reason);
}
}Features
- Confidence-Aware: Dynamically adjust permissions based on the AI agent's confidence level.
- Human-in-the-loop: Require human approval for low-confidence or high-risk actions.
- Audit Trails: SOC 2-ready logging of all agent decisions and actions.
- Framework Agnostic: Works with LangChain, AutoGPT, or custom agent frameworks.
Documentation
For full documentation, visit https://ui-seven-amber.vercel.app/.
License
MIT
