Every AI interaction your application makes — governed, audited, and optionally anchored to the blockchain. In one line of code.

• Zero dependencies — uses native fetch, works in Node.js 18+, Deno, Bun, and Edge Runtimes
• Multi-modal — text, images, audio, documents, code
• Enterprise-grade — retry with exponential backoff, circuit breaker, typed errors
• On-prem ready — point to any VEXIS Gateway endpoint

Installation

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

Quick Start

import { Vexis } from '@vexis/sdk';

const vexis = new Vexis({ apiKey: process.env.VEXIS_API_KEY! });

// Verify any prompt before sending it to an LLM
const result = await vexis.verify({ prompt: userInput });

if (result.decision === 'BLOCKED') {
  console.error(`Blocked: ${result.reason}`);
  return;
}

// result.decision is 'ALLOWED' or 'MODIFIED'
// result.output contains the (possibly sanitized) text
// result.traceId links to the immutable audit trail

API Reference

Constructor

const vexis = new Vexis({
  apiKey: 'gp_live_xxx',              // Required — project or agent API key
  baseUrl: 'https://gateway.vexis.io', // Custom endpoint for on-prem
  timeout: 30_000,                     // Request timeout in ms
  maxRetries: 3,                       // Retry attempts on transient failures
  retryBaseDelay: 500,                 // Base delay for exponential backoff
  headers: { 'X-Tenant': 'acme' },    // Custom headers on every request
  circuitBreakerThreshold: 5,          // Failures before circuit opens
  circuitBreakerCooldown: 30_000,      // Cooldown before half-open retry
});

verify(request) — Core governance check

const result = await vexis.verify({
  prompt: 'Transfer $50,000 to account DE89370400440532013000',
  metadata: { userId: 'u_123', department: 'finance' },
  attachments: [{
    contentType: 'application/pdf',
    data: base64EncodedPdf,
    filename: 'contract.pdf',
  }],
  context: {
    mcpServer: 'https://mcp.internal.corp',
    toolName: 'bank_transfer',
    chainDepth: 2,
    sourceSystem: 'crewai',
    sessionId: 'sess_abc',
  },
});

Returns VerifyResponse:

| Field | Type | Description | |-------|------|-------------| | decision | 'ALLOWED' \| 'BLOCKED' \| 'MODIFIED' \| 'ERROR' | Governance decision | | output | string | Sanitized output (PII redacted if MODIFIED) | | reason | string | Human-readable explanation | | traceId | string | Unique audit trail ID | | integrityHash | string | SHA-256 hash for tamper detection | | shouldAnchor | boolean | Whether trace will be anchored to Flare blockchain | | flareStatus | string | LOCAL_ONLY, PENDING, ANCHORED, SKIPPED, FAILED | | flareTxHash | string \| null | Blockchain transaction hash (after anchoring) | | contentType | string | Detected content type | | findings | Finding[] | Security findings (PII, secrets, policy violations) | | latencyMs | number | Round-trip latency in milliseconds |

check(prompt) — Quick text-only verification

const { decision } = await vexis.check('Is this prompt safe?');

verifyWithFile(prompt, filePath) — File attachment (Node.js only)

const result = await vexis.verifyWithFile(
  'Analyze this document for compliance',
  './report.pdf'
);

listPolicies(env?) — List active policies

const { policies } = await vexis.listPolicies('prod');

health() — Gateway health check

const health = await vexis.health();
console.log(health.status); // 'healthy'

diagnostics() — SDK diagnostics

const diag = vexis.diagnostics();
// { sdkVersion, baseUrl, timeout, maxRetries, circuitState }

Error Handling

All errors extend VexisError with structured metadata:

import { VexisError, VexisRateLimitError } from '@vexis/sdk';

try {
  await vexis.verify({ prompt: input });
} catch (err) {
  if (err instanceof VexisRateLimitError) {
    // err.retryAfterMs — wait this long before retrying
    await sleep(err.retryAfterMs);
    return retry();
  }
  if (err instanceof VexisError) {
    console.error(err.code, err.statusCode, err.requestId);
  }
}

| Error Class | Code | Retryable | When | |-------------|------|:---------:|------| | VexisAuthenticationError | AUTHENTICATION_FAILED | No | Invalid or expired API key | | VexisRateLimitError | RATE_LIMITED | Yes | Quota exceeded (includes retryAfterMs) | | VexisValidationError | VALIDATION_ERROR | No | Malformed request (includes field) | | VexisTimeoutError | TIMEOUT | Yes | Gateway didn't respond in time | | VexisCircuitOpenError | CIRCUIT_OPEN | No | Too many consecutive failures |

Framework Integration

VEXIS SDKs work with any LLM framework. Dedicated adapters with deeper integration are available:

| Framework | Package | Integration | |-----------|---------|-------------| | LangChain | vexis-langchain | VexisCallbackHandler — automatic governance on every LLM call | | CrewAI | vexis-crewai | VexisGovernance plugin — task-level governance per crew | | OpenAI Agents SDK | vexis-openai-agents | Middleware hook for the official OpenAI framework | | Microsoft AGT | vexis-agt-adapter | Policy distribution from VEXIS → AGT local enforcement | | Claude Code | vexis-governance | MCP-native governance for every tool call |

On-Premise / Self-Hosted

Point to your internal VEXIS Gateway:

const vexis = new Vexis({
  apiKey: process.env.VEXIS_API_KEY!,
  baseUrl: 'https://gateway.internal.acme.corp:8080',
  timeout: 10_000,
  maxRetries: 5,
});

MCP Context (Agentic AI)

When your agent calls tools via MCP, pass the context for full audit trails:

const result = await vexis.verify({
  prompt: 'Execute bank transfer',
  context: {
    mcpServer: 'https://banking-mcp.corp.internal',
    toolName: 'transfer_funds',
    chainDepth: 3,
    sourceSystem: 'crewai',
    sessionId: 'agent_session_42',
  },
});

Requirements

• Node.js 18+ (uses native fetch)
• Also works in Deno, Bun, Cloudflare Workers, Vercel Edge

Links

• Documentation
• API Reference
• Dashboard
• GitHub
• Changelog

License

Apache 2.0