LangChain

LangChain reaches OpenAI-compatible services through its OpenAI integration: ChatOpenAI from langchain-openai in Python and from @langchain/openai in TypeScript. Point it at our base URL and pass your cheapaimodel key.

Set it up

The table uses the Python names. In TypeScript they are configuration.baseURL, apiKey and model: the base URL goes inside configuration.

pip install -U langchain-openai

import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="deepseek-v4-1-flash",
    base_url="https://api.cheapaimodel.ai/v1",
    api_key=os.environ["CHEAPAIMODEL_API_KEY"],
)

response = llm.invoke("Explain recursion in one sentence.")
print(response.text)

# Streaming
for chunk in llm.stream("Write a haiku about the sea."):
    print(chunk.text, end="", flush=True)

ChatOpenAI sends requests to /v1/chat/completions. Any model in the catalog that lists that endpoint works in place of deepseek-v4-1-flash.

With init_chat_model

In Python, init_chat_model from the langchain package takes the same settings. Pass model_provider="openai": without it, LangChain guesses the provider from the model ID and picks the wrong integration, or none.

import os
from langchain.chat_models import init_chat_model

llm = init_chat_model(
    "deepseek-v4-1-flash",
    model_provider="openai",
    base_url="https://api.cheapaimodel.ai/v1",
    api_key=os.environ["CHEAPAIMODEL_API_KEY"],
)

Embeddings

cheapaimodel has no embeddings models yet. Keep OpenAIEmbeddings, or whichever embeddings class your retriever uses, on its own provider, and don't point it at our base URL.

Common mistakes

• 'baseURL' does not exist in type 'ChatOpenAIFields' (TypeScript). baseURL belongs inside configuration. Plain JavaScript ignores the top-level field without an error, and the request goes to OpenAI with your cheapaimodel key.
• requires the langchain-deepseek package or Unable to infer model provider from init_chat_model. Add model_provider="openai".
• 503 with model_not_found on /v1/embeddings. OPENAI_BASE_URL or OPENAI_API_BASE is exported with our base URL. ChatOpenAI reads it, but so does OpenAIEmbeddings, which then asks us for an embeddings model. Pass base_url in code instead of exporting it.
• <bound method BaseMessage.text ...> is printed. Python is older than 3.10, so pip installed the 0.3 line, where text is a method. Upgrade Python and reinstall.
• 503 with model_not_found on /v1/chat/completions. model is missing, so LangChain sends its OpenAI default gpt-3.5-turbo, or the ID isn't in the catalog. Set model to an ID from the catalog.

For base URL mistakes such as a missing /v1, see Base URL.