@arelis-ai/governance-providers-huggingface

Hugging Face Inference API + Dedicated Endpoints provider adapter for Arelis.

Install

pnpm add @arelis-ai/governance-providers-huggingface

Usage (Inference API)

import {
  createArelisClient,
  createModelRegistry,
  createAllowAllEngine,
  createConsoleSink,
} from '@arelis-ai/arelis-governance-sdk';
import { HuggingFaceProvider } from '@arelis-ai/governance-providers-huggingface';

const provider = new HuggingFaceProvider(
  {
    token: process.env.HF_TOKEN,
    apiType: 'inference',
  },
  ['meta-llama/Llama-3-8b-instruct']
);

const registry = createModelRegistry();
registry.register(provider);

const client = createArelisClient({
  modelRegistry: registry,
  policyEngine: createAllowAllEngine(),
  auditSink: createConsoleSink(),
});

const result = await client.models.generate({
  model: 'meta-llama/Llama-3-8b-instruct',
  request: {
    model: 'meta-llama/Llama-3-8b-instruct',
    messages: [{ role: 'user', content: 'Hello HF' }],
  },
  context: {
    org: { id: 'org_1' },
    actor: { type: 'human', id: 'user_1' },
    purpose: 'testing',
    environment: 'dev',
  },
});

console.log(result.output.content);

Usage (OpenAI-compatible endpoint)

const provider = new HuggingFaceProvider(
  {
    token: process.env.HF_TOKEN,
    endpoint: 'https://your-endpoint.huggingface.cloud',
    apiType: 'openai',
  },
  ['tgi-openai']
);

Notes

• apiType: 'openai' enables streaming, tool calls, and multimodal inputs.
• apiType: 'inference' uses the standard Inference API and returns text only.