Lite Agent
  • English

      • Supported providers

      Direct adapters and the AI SDK 7 bridge feed the same createLiteAgent() entry point. The current provider package requires Node.js >=22.

      Choose an integration path

      ScenarioRecommended entry
      OpenAI Chat Completions / Anthropic Messagesopenai() / anthropic() from @lite-agent/provider
      Local Ollama, vLLM, LM Studio or llama.cpplocalOpenAI() from the same package; the service must expose Chat Completions compatibility
      OpenAI Responses, Gemini, cloud platforms and other vendorsaiSdk(vendorModel), with an adapter installed as needed
      A vendor or private gateway exposes Chat Completions compatibilityopenai({ baseURL }) or AI SDK createOpenAICompatible()

      Check the protocol matrix, then select an adapter below. See AI SDK integration and direct adapter configuration for complete examples, authentication and parameters.

      Providers and adapters

      These paths have offline HTTP protocol tests using the actual vendor SDKs. This is not account-level or all-model certification. Coverage comprises 21 vendors/platforms, both OpenAI protocols and one generic compatible path.

      Vendor/platformInstall as neededModel factoryTested protocol
      OpenAI@ai-sdk/openaicreateOpenAI()(modelId)Responses / Chat
      Anthropic@ai-sdk/anthropiccreateAnthropic()(modelId)Messages
      Google Gemini@ai-sdk/googlecreateGoogle()(modelId)Gemini
      Azure OpenAI@ai-sdk/azurecreateAzure(settings)(deploymentId)Responses
      Amazon Bedrock@ai-sdk/amazon-bedrockcreateAmazonBedrock(settings)(modelId)Converse
      Google Vertex@ai-sdk/google-vertexcreateGoogleVertex(settings)(modelId)Vertex Gemini
      DeepSeek@ai-sdk/deepseekcreateDeepSeek()(modelId)Chat
      Alibaba Qwen / 千问@ai-sdk/alibabacreateAlibaba()(modelId)Chat
      Z.AI GLM / 智谱@ai-sdk/zaicreateZai()(modelId)Chat
      Moonshot Kimi@ai-sdk/moonshotaicreateMoonshotAI()(modelId)Chat
      MiniMax@ai-sdk/minimaxcreateMiniMax()(modelId)Anthropic-compatible
      xAI Grok@ai-sdk/xaicreateXai()(modelId)Responses
      Mistral@ai-sdk/mistralcreateMistral()(modelId)Chat
      Groq@ai-sdk/groqcreateGroq()(modelId)Chat
      Cohere@ai-sdk/coherecreateCohere()(modelId)Cohere v2
      Together AI@ai-sdk/togetheraicreateTogetherAI()(modelId)Chat
      Fireworks@ai-sdk/fireworkscreateFireworks()(modelId)Chat
      Cerebras@ai-sdk/cerebrascreateCerebras()(modelId)Chat
      DeepInfra@ai-sdk/deepinfracreateDeepInfra()(modelId)Chat
      Perplexity@ai-sdk/perplexitycreatePerplexity()(modelId)Chat (text-only verification)
      OpenRouter@openrouter/ai-sdk-providercreateOpenRouter()(modelId)Chat (community adapter)

      Configure keys, endpoints, regions, projects, credential chains and custom fetch in the vendor's createXxx(settings) factory. Azure uses deployment ids. Bedrock and Vertex retain their official authentication mechanisms.

      Perplexity is covered as a text provider; configure allowedTools: [], not native tool calls. Other rows validate a protocol path, not every available model's tool capability. OpenRouter is a community adapter maintained by its team.

      Compatible endpoints: Doubao, Baidu, Tencent, SiliconFlow and private endpoints

      For an endpoint explicitly documented as OpenAI Chat compatible, use the maintained compatibility adapter:

      import { aiSdk } from '@lite-agent/provider';
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';

const compatible = createOpenAICompatible({
  name: 'my-endpoint',
  baseURL: process.env.MODEL_BASE_URL!, // Vendor-documented URL for your region/product
  apiKey: process.env.MODEL_API_KEY,
});
const provider = aiSdk(compatible(process.env.MODEL_ID!));

      This validates the common protocol bridge, not each remote deployment or account. Image/video-only provider packages are not language models. Community adapters must expose V4. Existing openai(), anthropic() and localOpenAI() remain available.

      Validation and capability limits

      StatusMeaning
      Offline wire verificationActual vendor SDKs run against response fixtures for tool replay, streaming, usage, errors and no automatic retry; Perplexity is text-only
      Generic compatibility verificationThe compatible adapter is tested; individual Doubao, Baidu, Tencent, SiliconFlow or private deployments still need endpoint/model validation
      Live model verificationugreen-ai-model completed two model requests and one file-tool call through the bridge; this does not certify every vendor's live service

      Vendor support does not imply identical capabilities across all its models. Verify tool calling, reasoning effort, context window, region and account access for your model. Configure allowedTools: [] for Perplexity. The bridge currently handles text, reasoning and function tools, not image/audio/video generation or provider-executed tools.

      Community adapters must implement current LanguageModelV4; V2/V3 are unsupported. Additional compatible vendors do not require a new agent type.