@lanlinling/llm-providers

Multi-provider LLM chat-completion adapter registry — OpenAI, Anthropic, Google, Azure, Amazon Bedrock, and every OpenAI-compatible provider (DeepSeek, Qwen, Kimi, MiniMax, GLM, SiliconFlow, Doubao, Grok, Ollama, and more). Extracted from OpenMAIC.

Also includes onehub — a personal self-hosted aggregator gateway (one key, ~180 models across vendors including Claude/GPT/Gemini/DeepSeek). See TROUBLESHOOTING.md for why its model IDs and capabilities (no extended thinking) differ from the native provider entries.

Install

{
  "dependencies": {
    "@lanlinling/llm-providers": "github:bluemomo112/ai-providers#path:packages/llm-providers"
  }
}

Add it to your project's pnpm-workspace.yaml (pnpm blocks git-dependency build scripts by default):

onlyBuiltDependencies:
  - "@lanlinling/llm-providers"

Does it read API keys from my environment?

No. Nothing in this package touches process.env for credentials (the only process.env reads are AWS's own AWS_REGION/BEDROCK_REGION convention for Bedrock, and log-level config). You always pass apiKey explicitly — this is deliberate, so the package never silently depends on an env var name you didn't choose.

If you want the common "load from .env and pass it in" pattern:

import 'dotenv/config'; // or your own .env loader
import { getModel } from '@lanlinling/llm-providers';
import { generateText } from 'ai';

const { model } = getModel({
  providerId: 'anthropic',
  modelId: 'claude-sonnet-4-6',
  apiKey: process.env.ANTHROPIC_API_KEY!,
  baseUrl: process.env.ANTHROPIC_BASE_URL, // optional, provider default otherwise
});

const result = await generateText({ model, prompt: 'Reply with exactly one word: pong' });
console.log(result.text);

This is exactly the call this package was smoke-tested with (real Anthropic API call, live response).

Usage

import { getModel, getAllModels, getProvider, PROVIDERS } from '@lanlinling/llm-providers';

// List every built-in provider + its models
for (const { provider, models } of getAllModels()) {
  console.log(provider.name, models.map((m) => m.id));
}

// Look up one provider's metadata (default base URL, whether it needs a key, etc.)
console.log(getProvider('deepseek'));

// Get a ready-to-use Vercel AI SDK LanguageModel for any provider
const { model, modelInfo } = getModel({
  providerId: 'deepseek',
  modelId: 'deepseek-chat',
  apiKey: process.env.DEEPSEEK_API_KEY!,
});

PROVIDERS is a plain mutable object — extend it at startup if you need a provider that isn't built in, or just pass providerType explicitly in getModel()'s config to skip the registry lookup entirely.

See TROUBLESHOOTING.md for provider-specific gotchas (proxy base-URL formats, custom providers, etc.).