[feature request] Allow passing custom HTTP headers to langchain_ollama.ChatOllama(...) and pass ChatOllama(api_key=) to the Authorization: Bearer ... header)

[vadimkantorov]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Classes like ChatOpenAI/ChatGoogleGenerativeAI support accepting api_key= constructor argument and pass it into appropriate HTTP header.

Vanilla Ollama server currently does not support checking an API key from HTTP headers, but people still roll custom nginx wrapper proxies.

I propose to still allow passing extra headers= dict argument (allowing to pass in dict(Authorization = f"Bearer {my_custom_api_key}"))

I noticed that ChatOllama(...) does not throw if I pass in api_key=. But it appears that is simply ignored and is not passed to any HTTP header: https://github.com/langchain-ai/langchain/blob/master/libs/partners/ollama/langchain_ollama/chat_models.py

For comparison, for Mistral, Authorization header is being constructed out of self.mistral_api_key: https://github.com/langchain-ai/langchain/blob/master/libs/partners/mistralai/langchain_mistralai/chat_models.py#L526-L536

Ollama's Client class actually supports arguments headers=: https://github.com/ollama/ollama-python/blob/main/ollama/_client.py#L73

So ChatOllama could just enable passing it through to the underlying Client / AsyncClient in _set_clients

Thanks!

Motivation

Basic protection of publicly exposed Ollama server

Proposal (If applicable)

No response

[mdrxy]

While ChatOllama doesn't have a direct headers or api_key constructor argument, you can pass custom headers using the client_kwargs parameter. This arg accepts a dictionary of keyword arguments that are passed directly to the underlying httpx HTTP client used by LangChain.

To add an authorization header, you would pass a dictionary to client_kwargs that contains a headers key, e.g.:

from langchain_ollama import ChatOllama

my_custom_api_key = "your-secret-api-key"

llm = ChatOllama(
    model="llama3.1",
    # Pass custom headers to the underlying httpx client
    client_kwargs={
        "headers": {
            "Authorization": f"Bearer {my_custom_api_key}"
        }
    }
)

# Now, any request made by this llm instance will include the specified header.
# llm.invoke("your prompt here")