@sentinos/node

Official TypeScript-first SDK for Sentinos. Use it to govern agent actions, inspect decision traces, replay policy outcomes, and connect Node runtimes to Sentinos services from one typed client.

Requirements

• Node.js 20 or newer (ESM runtime only)
• The package is shipped as an ES module, so your loader/build tooling must honor type: "module"

Docs & resources

• Sentinos SDK docs
• Repository and API surface
• Release notes
• Issue tracker

Quickstart

npm install @sentinos/node

Before you run the first request, make sure you have:

• a Sentinos API base URL such as https://api.sentinos.ai
• an organization id
• an access token or API key

import { JWTAuth, SentinosClient } from "@sentinos/node";

const client = SentinosClient.simple("https://api.sentinos.ai", {
  orgId: "acme",
  auth: new JWTAuth(async () => process.env.SENTINOS_ACCESS_TOKEN ?? ""),
});

const result = await client.kernel.execute({
  tenant_id: "acme",
  agent_id: "assistant-1",
  session_id: "sess-1",
  intent: {
    type: "tool_call",
    tool: "stripe.refund",
    args: { amount: 1200 },
  },
});

console.log(result.decision, result.trace_id);

SentinosClient.simple wires up the base URL, organization, and authentication so you can focus on orchestrating agents.

Environment configuration

import { SentinosClient } from "@sentinos/node";

const client = SentinosClient.fromEnv();

The SDK honors the following environment variables:

• SENTINOS_BASE_URL (aliases: SENTINOS_API_URL, SENTINOS_URL)
• SENTINOS_KERNEL_URL
• SENTINOS_ARBITER_URL
• SENTINOS_CHRONOS_URL
• SENTINOS_CONTROLPLANE_URL
• SENTINOS_MESHGATE_URL
• SENTINOS_ORG_ID (alias: SENTINOS_TENANT_ID)
• SENTINOS_ACCESS_TOKEN
• SENTINOS_TIMEOUT_SECONDS

Workforce sessions add:

• SENTINOS_WORKFORCE_IDP_ISSUER
• SENTINOS_WORKFORCE_EXCHANGE_AUDIENCE
• SENTINOS_WORKFORCE_REQUESTED_TTL_MINUTES

Authentication

JWT

import { JWTAuth, SentinosClient } from "@sentinos/node";

const client = SentinosClient.fromEnv({
  orgId: "acme",
  auth: new JWTAuth(async () => getFreshOidcToken()),
});

API Key

import { APIKeyAuth, SentinosClient } from "@sentinos/node";

const client = new SentinosClient({
  orgId: "acme",
  kernelUrl: process.env.SENTINOS_KERNEL_URL!,
  arbiterUrl: process.env.SENTINOS_ARBITER_URL!,
  chronosUrl: process.env.SENTINOS_CHRONOS_URL!,
  controlplaneUrl: process.env.SENTINOS_CONTROLPLANE_URL!,
  auth: new APIKeyAuth(process.env.SENTINOS_API_KEY ?? ""),
});

Workforce tokens

import { SentinosClient, WorkforceTokenProvider } from "@sentinos/node";

const workforceAuth = WorkforceTokenProvider.fromEnv(async () => ({
  externalSubject: "user-123",
  email: "[email protected]",
  displayName: "Example User",
  groups: ["finance-reviewers"],
  assertionToken: process.env.IDP_ASSERTION,
}));

const client = SentinosClient.fromEnv({ auth: workforceAuth });

Trace & replay workflow

Inspecting a trace and replaying it lets you verify why a decision landed where it did and capture evidence for auditors.

const trace = await client.traces.getTrace("trace_123");

const replay = await client.traces.replayTrace(trace.trace_id, {
  include_explain: true,
  environment_assumptions: { user_tier: "trial" },
});

console.log("replayed decision", replay.decision);

If you need a full forensic package, follow up with client.traces.exportReplayEvidence(...) or client.traces.replayTraceMatrix(...) to generate evidence-ready replay output.

Runtime integrations

The SDK bundles LLMGuard plus adapters for OpenAI chat completions, OpenAI responses, OpenRouter, and Anthropic so you can anchor guardrails around any completion stream.

import { createOpenAIResponsesAdapter, LLMGuard, SentinosClient } from "@sentinos/node";

const client = SentinosClient.fromEnv();
const guard = new LLMGuard({
  kernel: client.kernel,
  agentId: "assistant-1",
  sessionId: "sess-42",
});

const adapter = createOpenAIResponsesAdapter({ guard, client: openai });

const result = await adapter.create({
  model: "gpt-4.1-mini",
  input: [{ role: "user", content: "summarize incidents" }],
});

console.log(result.trace.trace_id, result.trace.decision);

Client surface

• client.kernel for governed execution and runtime operations
• client.traces for trace lookup, replay, export, cost, and lineage
• client.arbiter for policy lifecycle and simulation
• client.chronos for context snapshots and provenance
• client.alerts, client.incidents, client.marketplace, and client.meshgate for adjacent Sentinos workflows