@interven/sdk

TypeScript / JavaScript SDK for the Interven AI firewall. Scan agent tool calls before they execute — block malicious requests, redact PII and secrets, and route risky actions to human approval.

npm install @interven/sdk

Quickstart

import { Client } from "@interven/sdk";

const client = new Client({ apiKey: process.env.INTERVEN_API_KEY });

const result = await client.scan({
  method: "POST",
  url: "https://slack.com/api/chat.postMessage",
  body: { text: "Customer SSN 478-23-9156, email [email protected]" },
});

if (result.decision === "ALLOW") {
  await sendToSlack(originalBody);
} else if (result.decision === "SANITIZE") {
  await sendToSlack(result.sanitizedBody);
} else if (result.decision === "REQUIRE_APPROVAL") {
  await pollApproval(result.approvalId!);
} else {
  console.warn("Blocked by Interven:", result.reasonCodes);
}

That's it. Get an API key at intervensecurity.com (free tier: 1,000 scans/month).

Decisions

| Decision | What to do | |----------|-----------| | ALLOW | Forward the original request to the upstream API | | DENY | Block the call. Reason codes explain why. | | SANITIZE | Forward result.sanitizedBody instead of the original — secrets/PII have been redacted | | REQUIRE_APPROVAL | Pause; poll /approvals/{id}/status until decided |

Configuration

| Option | Env var | Default | |--------|---------|---------| | apiKey | INTERVEN_API_KEY | — (required) | | gatewayUrl | INTERVEN_GATEWAY_URL | https://api.intervensecurity.com | | timeoutMs | — | 30000 | | agentId | — | unset (server uses default) | | runtimeType | — | "node" |

Framework recipes

LangChain.js

import { Client } from "@interven/sdk";
import type { ToolRunnableConfig } from "@langchain/core/tools";

const interven = new Client({ apiKey: process.env.INTERVEN_API_KEY });

async function safeFetch(url: string, init: RequestInit) {
  const result = await interven.scan({
    method: (init.method as string) ?? "GET",
    url,
    body: init.body ? JSON.parse(init.body as string) : {},
    runtimeType: "langchain",
  });
  if (result.decision === "DENY") throw new Error(`Blocked: ${result.reasonCodes.join(", ")}`);
  return fetch(url, init);
}

Generic agent / MCP server

Wrap any outbound HTTP call with client.scan(...) before sending. Works with the Vercel AI SDK, OpenAI Agents, custom MCP servers, etc.

Legacy: HMAC AifClient

The original HMAC-signed /invoke flow is still supported for existing customers:

import { AifClient } from "@interven/sdk";

const client = new AifClient({
  gatewayUrl: "http://localhost:4000",
  agentId: "00000000-0000-0000-0000-000000000010",
  agentName: "release-bot",
  agentSecret: process.env.AGENT_SECRET!,
});

const r = await client.invoke({
  toolName: "github",
  method: "PUT",
  urlPath: "/repos/acme/main-app/collaborators/external-user",
  credentialType: "pat",
  credentialToken: process.env.GH_TOKEN!,
  scopes: ["repo"],
});

New integrations should prefer Client — fewer required fields and no secret to manage.

Errors

import {
  IntervenAuthenticationError,
  IntervenGatewayError,
  IntervenPayloadTooLargeError,
} from "@interven/sdk";

try {
  await client.scan({ method: "GET", url: "..." });
} catch (e) {
  if (e instanceof IntervenAuthenticationError) { /* bad key */ }
  if (e instanceof IntervenPayloadTooLargeError) { /* >256KB */ }
  if (e instanceof IntervenGatewayError) { /* network or 5xx */ }
}

License

MIT © Interven Security

Links

• Interven docs
• API reference
• Get an API key
• GitHub