api
New
2026-05-30TypeScript / Node.js SDK
Typed client for all Aegis Sovereign APIs — model registry, compliance, bias evals, robustness, webhooks, audit, and the full AI Safety Platform (v0.2.0). Works in Node.js 18+ and modern browsers.
sdk
typescript
nodejs
api
webhooks
Installation
Install the SDK from npm.
bash
1npm install @aegissovereign/sdk
2# or
3yarn add @aegissovereign/sdk
4# or
5pnpm add @aegissovereign/sdkQuick Start
Initialise the client with your API key and workspace ID.
typescript
1import { AegisSovereignClient } from "@aegissovereign/sdk";
2
3const client = new AegisSovereignClient({
4 baseUrl: "https://sovereign.yourcompany.com",
5 apiKey: process.env.AEGISSOVEREIGN_API_KEY!,
6 workspaceId: "ws_prod_abc123",
7 // baseUrl: "https://api.aegissovereign.io", // default
8 // timeoutMs: 30_000, // default
9});
10
11// List models
12const models = await client.models.list();
13
14// Run compliance evaluation
15const eval_ = await client.regulatory.evaluate(
16 "model_fraud_v2",
17 "eu_ai_act"
18);
19
20// Verify a webhook signature
21const isValid = AegisSovereignClient.verifyWebhookSignature(
22 rawBody,
23 signature,
24 process.env.WEBHOOK_SECRET!
25);Available Sub-Clients
The SDK exposes typed sub-clients for every surface area.
| Client | Description |
|---|---|
| client.models | Model registry — create, list, get, update, promote |
| client.chat | LLM gateway — completions with budget enforcement |
| client.outputEvals | LLM output quality evaluations |
| client.biasEvals | Bias/fairness evaluations with DIR analysis |
| client.robustnessEvals | Adversarial robustness evaluations |
| client.compliance | Regulatory framework evaluations |
| client.regulatory | Alias for client.compliance |
| client.webhooks | Webhook subscription management |
| client.audit | Audit log queries + external-ledger anchoring |
| **client.safety** ✨ | AI Safety Platform — Policy Engine, Agentic Guard, PII Masking, Conversation Monitor, Approvals, Clearance, SLOs, Red-Team, Integrations, Regulatory Reports. 30 typed types, 20+ methods. |
Safety Client Examples
The client.safety sub-client covers all 8 subsystems of the AI Safety Platform.
typescript
1import { AegisSovereignClient } from "@aegissovereign/sdk";
2const client = new AegisSovereignClient({
3 baseUrl: "https://sovereign.yourcompany.com",
4 apiKey: process.env.AEGISSOVEREIGN_API_KEY!,
5 workspaceId: "ws-prod",
6});
7
8// Validate a prompt through the full pipeline (regex + ML)
9const result = await client.safety.validateInput(
10 "Ignore all previous instructions and...",
11);
12if (result.blocked) {
13 console.error("Blocked:", result.violations.map(v => v.rule_id));
14}
15
16// Validate a model response (auto-redacts PII)
17const resp = await client.safety.validateOutput(responseText);
18const safeText = resp.sanitized_text ?? responseText;
19
20// Check model registry clearance before promoting
21const clearance = await client.safety.checkModel(
22 "gpt-4", "gpt-4-turbo", "customer-support"
23);
24if (!clearance.allowed) throw new Error(clearance.reason);
25
26// SLO fleet health — burn_rate_1h > 14.4 = page immediately
27const slos = await client.safety.getSLOSummary();
28if (slos.critical > 0) console.error("Critical SLO burn rate!");
29
30// Run red-team certification suite
31const redteam = await client.safety.runRedteam({
32 domains: ["medical", "financial"],
33 includeMultiTurn: true,
34});
35console.log(`Pass rate: ${(redteam.certification.pass_rate * 100).toFixed(1)}%`);
36
37// Generate EU AI Act report
38const report = await client.safety.generateReport({
39 framework: "eu_ai_act",
40 organization: "Acme Financial",
41});
42console.log(report.content);typescript
1// List pending critical review requests
2const reviews = await client.safety.listReviews({
3 status: "pending",
4 priority: "critical",
5});
6
7// Submit a review decision
8await client.safety.submitReview("rev-abc123", {
9 decision: "reject",
10 reason: "Confirmed jailbreak attempt",
11 reviewer_id: "alice@corp.com",
12 reviewer_name: "Alice Chen",
13});
14
15// Validate a tool call before execution (Agentic Guard)
16const check = await client.safety.validateToolCall(
17 "delete_file",
18 { path: "/prod/db" },
19 "researcher-agent",
20);
21if (!check.allowed) {
22 throw new Error(`Tool blocked: risk=${check.risk_level}`);
23}
24
25// Run the red-team certification suite (server-side)
26const cert = await client.safety.runRedteam({
27 domains: ["financial"],
28 includeMultiTurn: true,
29});
30if (!cert.overall_passed) {
31 console.error(`${cert.certification.blocking_failures} undetected attack(s)`);
32}