@datavibe/sdk

Official TypeScript/JavaScript SDK for DataVibe — AI Execution Security Gateway.

DataVibe intercepts AI-generated outbound content before it reaches customers. Policy scan → human approval queue → tamper-evident audit log. One API call.

Architecture

Your LLM API keys never leave your infrastructure. DataVibe only receives the text your model produced.

Your system:
  const output = await openai.chat.completions.create({ ... })
                 ↑ your key, your model, your network

  const verdict = await datavibe.check({ content: output.choices[0].message.content })
                  ↑ only the generated text reaches DataVibe

Install

npm install @datavibe/sdk
# or
pnpm add @datavibe/sdk
# or
yarn add @datavibe/sdk

Quickstart

import { DataVibeClient } from "@datavibe/sdk";

const datavibe = new DataVibeClient({
  apiKey: process.env.DATAVIBE_API_KEY!, // dv_live_... from your dashboard
});

Get your API key at app.datavibe.cc/settings/api-keys.

Three APIs

check() — Governance verdict on any AI output

You call your LLM. Send only the output to DataVibe. Returns safe | blocked | review_required.

const aiOutput = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Write a sales email to Acme Corp." }],
});

const verdict = await datavibe.check({
  content: aiOutput.choices[0].message.content,
  contentType: "email",
  sourceModel: "gpt-4o", // model name only — never an API key
});

if (verdict.verdict === "safe") {
  await sendEmail(aiOutput.choices[0].message.content);
} else if (verdict.verdict === "review_required") {
  // Route to human reviewer
  console.log("Review at:", verdict.review_url);
} else {
  // "blocked" — hard policy violation, do not send
  console.log("Blocked by:", verdict.violations.map((v) => v.rule));
}

intercept() — Full outbound email governance workflow

Submits an AI-generated email for policy scan → human approval queue → dispatch.

const result = await datavibe.intercept({
  recipient: prospect.email,
  subject: aiSubject,
  body_html: aiGeneratedEmail,
  source_model: "claude-3-5-sonnet-20241022",
  metadata: {
    deal_id: deal.salesforceId,
    rep_id: rep.id,
  },
});

if (result.status === "BLOCKED") {
  // Hard policy violation — never dispatched
  await notifyRep(result.policy_violations);
} else if (result.status === "QUEUED") {
  // Routed to human approval queue
  await notifyReviewer(result.review_url);
} else {
  // SENT — all controls passed
  console.log("Dispatched, action_id:", result.action_id);
}

generateAndCheck() — Single-call generate + govern

DataVibe calls the LLM (shared key or your BYOK) and governance-checks the output in one round trip.

const result = await datavibe.generateAndCheck({
  messages: [{ role: "user", content: "Write a sales email to Acme Corp." }],
  contentType: "email",
  model: "gpt-4o-mini",
});

if (result.verdict === "safe") {
  await sendEmail(result.content!);
} else if (result.verdict === "review_required") {
  await routeToApprovalQueue(result.review_url);
} else {
  // blocked — result.content is null
  console.log("Blocked by:", result.violations.map((v) => v.rule));
}

LangChain Integration

Callback Handler (recommended)

Automatically governance-checks every LLM output in your chain — no manual wrapping required.

import { DataVibeClient } from "@datavibe/sdk";
import { DataVibeCallbackHandler } from "@datavibe/sdk/integrations";
import { ChatOpenAI } from "@langchain/openai";
import { HumanMessage } from "@langchain/core/messages";

const datavibe = new DataVibeClient({
  apiKey: process.env.DATAVIBE_API_KEY!,
});

const handler = new DataVibeCallbackHandler(datavibe, {
  contentType: "email",
  onBlocked: (verdict) => {
    console.error("LLM output blocked:", verdict.violations[0]?.rule);
  },
});

const llm = new ChatOpenAI({ model: "gpt-4o" });
const response = await llm.invoke(
  [new HumanMessage("Write a sales email to Acme Corp.")],
  { callbacks: [handler] },
);

Tool Wrapper

Wrap any LangChain tool with a governance check:

import { wrapTool } from "@datavibe/sdk/integrations";

const governedTool = wrapTool(datavibe, emailTool, {
  contentType: "email",
  sourceModel: "gpt-4o",
  onBlocked: (verdict) => `[blocked: ${verdict.violations[0]?.rule}]`,
  onReviewRequired: (verdict) => `Queued for review: ${verdict.review_url}`,
});

HIPAA / FINRA / EU AI Act compliance

DataVibe ships compliance policy packs for regulated industries. See datavibe.cc/governance/packages for HIPAA, FINRA, EU AI Act, and EEOC bundles.

// Healthcare example — intercept patient support bot reply before delivery
const result = await datavibe.intercept({
  recipient: patient.email,
  subject: "Re: Your appointment inquiry",
  body_html: botGeneratedReply,
  source_model: "claude-haiku-4-5",
  metadata: { patient_id: patient.id, bot_session: sessionId },
});

if (result.status === "BLOCKED") {
  // PHI detected — never sent, escalate to clinical ops
  await escalateToClinicalOps(patient.id, result.policy_violations);
}

Error handling

import { DataVibeGatewayTimeoutError } from "@datavibe/sdk";

try {
  const verdict = await datavibe.check({ content: aiOutput });
} catch (err) {
  if (err instanceof DataVibeGatewayTimeoutError) {
    // err.retryable === true — safe to retry with backoff
    console.error("Gateway timeout, retrying...");
  }
  throw err;
}

Configuration

| Option | Type | Default | Description | | ------------------ | -------- | -------------------------- | ----------------------------------------------------- | | apiKey | string | required | Workspace API key (dv_live_...) from your dashboard | | baseUrl | string | https://gate.datavibe.cc | Override the gateway URL | | gatewayTimeoutMs | number | 10000 | Request timeout in milliseconds |

Links

• Dashboard: app.datavibe.cc
• API docs: gate.datavibe.cc/docs
• Integration guide: datavibe.cc/docs/integration
• Status: datavibe.cc/status
• Python SDK: datavibe on PyPI

License

MIT © DataVibe