Z.ai integrations: third-party toolchain reference

The integration foundation: OpenAI compatibility

The BigModel API mirrors the OpenAI chat-completions contract, which means any tool that supports a custom OpenAI base URL can integrate Z.ai in minutes.

The BigModel API endpoint accepts requests formatted identically to OpenAI's chat-completions endpoint. The request body uses the same model, messages, temperature, and max_tokens fields. The response object follows the same structure, including the choices array and the usage token-count block. This compatibility means that the integration effort for most modern developer tools is limited to changing two values in a configuration file or settings screen: the base URL and the API key.

The GLM model identifiers used in the model field differ from OpenAI's identifiers — you use glm-4, glm-4-flash, or similar strings rather than gpt-4o. This is the only semantic change in the request object for standard chat-completions integrations. Tools that hard-code model names (rather than taking them from a configuration field) will need a minor code change at the call site.

Framework integrations

LangChain, LlamaIndex, and Haystack each have first-party or community-maintained support for Z.ai.

LangChain supports Z.ai through the ChatZhipuAI class in the langchain-community package. Instantiating it with a BigModel API key and a model name produces a chat model object that works identically to any other LangChain LLM — it can be dropped into chains, agents, and retrievers without modification. The class handles authentication, request formatting, and streaming responses internally. For teams already using LangChain with OpenAI, switching to Z.ai is a single class-name change at the model-constructor call.

LlamaIndex exposes Z.ai through an OpenAI-compatible LLM class with a custom base URL parameter. Setting OpenAI(base_url="https://open.bigmodel.cn/api/paas/v4/", api_key="...") in a LlamaIndex context routes all completions through BigModel. The model identifier needs to be a GLM name rather than a GPT name. All LlamaIndex abstractions — query engines, RAG pipelines, agent loops — work without modification once the LLM object is pointed at BigModel.

Haystack supports custom OpenAI-compatible providers through its OpenAIChatGenerator component, which accepts a api_base_url parameter. The integration pattern is the same as LlamaIndex: configure the base URL, supply the BigModel API key, and specify a GLM model name. Haystack pipelines built for OpenAI migrate to Z.ai with changes only to the generator component configuration.

Code editor integrations: Cursor and Continue.dev

Both Cursor and Continue.dev support custom model endpoints, allowing GLM models to power code completions and chat in the editor.

Cursor exposes model configuration through its settings panel under the Models section. Adding a custom model with an OpenAI-compatible API format, entering the BigModel base URL, and supplying a BigModel API key routes Cursor's completion and chat requests to the GLM family. The GLM code-specialised variants are appropriate for code completion workloads; the general-purpose GLM-4 tier handles the conversational Composer and Ask modes well.

Continue.dev uses a JSON configuration file (~/.continue/config.json) that accepts a model block with a provider of openai, a custom apiBase, and an apiKey. Setting these to the BigModel values adds a Z.ai model to the Continue model selector. Continue's autocomplete and chat features then use the configured GLM model. Community contributors have published pre-built Continue configurations for specific GLM variants that cover the recommended context window and temperature settings.

Ollama and automation tools

Ollama runs open-weight GLM builds locally; n8n and Zapier connect Z.ai to workflow automation pipelines.

Ollama serves open-weight models locally. For the GLM GGUF builds downloaded from community mirrors on Hugging Face, Ollama can load and serve them directly once the GGUF file is placed in the correct model directory. Community shims also exist that expose the BigModel closed-API models as Ollama-compatible endpoints for clients that prefer the Ollama interface. The shim approach is less stable than the direct BigModel API integration and works best for development rather than production use.

n8n's AI nodes include an OpenAI-compatible LLM node that accepts a custom base URL. Setting this to the BigModel endpoint and providing an API key makes GLM models available as AI action nodes in any n8n workflow. Zapier does not have a first-party Z.ai action, but the OpenAI action with a custom API configuration and the webhook action both serve as viable integration paths for teams that need GLM capability in Zapier automation sequences. Research on AI workflow orchestration from NIST is useful context for teams embedding AI models in production automation pipelines.

Z.ai integration patterns — five key tools, their integration method, and notes

Tool	Integration pattern	Notes
LangChain	ChatZhipuAI class in langchain-community	First-party support; drop-in replacement for ChatOpenAI in chains and agents
LlamaIndex	OpenAI(base_url=BigModel_URL, api_key=...)	Uses the OpenAI-compatible class; all LlamaIndex abstractions work unchanged
Cursor / Continue.dev	Custom model endpoint in settings or config.json	Enter BigModel base URL and API key; use GLM model identifier strings
Ollama	Direct GGUF loading for open-weight builds; community shim for closed-API models	Shim approach works for development; direct BigModel API preferred for production
n8n	OpenAI-compatible LLM node with custom base URL	Full node functionality available; no dedicated Z.ai node required