LibreChat always enforces a max context length of 4096

[pedrib]

I'm using the dockerized LibreChat, Ollama backend, with the cogito:32b model (supports up to 128k context length).

When I try to send a message with more than 4096 tokens, I get the following error message on the LibreChat web UI:

The latest message token count is too long, exceeding the token limit (13948 / 4095 respectively). Please shorten your message, adjust the max context size from the conversation parameters, or fork the conversation to continue.

Looking at the docker logs, I also see:

LibreChat         | 2025-04-15 10:30:16 warn: Prompt token count exceeds max token count (13952 / 4095).
LibreChat         | 2025-04-15 10:30:16 error: [handleAbortError] AI response error; aborting request: { "type": "INPUT_LENGTH", "info": "13952 / 4095"

I've tried editing librechat.yaml and adding the following:

custom:
  - name: Ollama
  (...)
  maxContextTokens: 128000
  max_tokens: 128000
  addParams:
    max_tokens: 128000
    maxContextTokens: 128000

But I still get the same error after restarting all the docker containers (with docker compose ...).
Using Ollama on the command line, the model seems to accept at least 32k tokens without a problem.

Am I doing something wrong here? What is the right way to increase the context length?

[pedrib]

A quick update:

I've added OLLAMA_CONTEXT_LENGTH: 16384 as an enviroment variable in docker-compose.override.yml and confirm that Ollama processes up to that context length via the command line (of course, memory usage increases substantially).

Also tried to modify librechat.yaml with the following:

custom:
  - name: Ollama
  (...)
  addParams:
    num_ctx: 16384

... but I still get the 4095 max context length message as shown above.

[pedrib]

Seems like the only way is to set up the context length in the UI, as described by @danny-avila here: #4259 (comment)

In the same thread he also mentions that in the future this will be a config file setting.

Closing this as my problem is solved, hope this helps someone in the future.