Setting max_tokens for OpenAI Compatible Provider for Aider

[scameronde]

I am completely lost and need help.

My situation:

• I am using AiderDesk on Linux as an AppImage
• when using OpenRouter, everything works perfect
• I am using Claude Haiku 4.5 for everything
• now I need to switch to another OpenAI compatible provider other than OpenRouter

My problem:

• the other provider requres that max_tokens is set with every call, or else only 4k output tokens are allowed
• I am too stupid to set max_tokens
• I have set all values in the model UI, but that seems to only have an affect on the Agent, not on Aider itself

I know that there are config files that can be passed to aider.
I know that there is a setting in AiderDesk for configuring Aider.

I just don't get all these loose ends together. I am afraid I need a step-by-step instruction.

Anyone out there being able to help? Thank you!

[wladimiiir]

In AiderDesk, it's currently only possible to set maxTokens (max output tokens) for Agent mode via the Agent profile settings.

If you need a specific settings for Aider, you will probably need to create .aider.model.metadata.json in your project root (or home) with the following content:

{
    "openai/name-of-your-model": {
        "max_tokens": 4096,
        "max_input_tokens": 32000,
        "max_output_tokens": 4096,
        "input_cost_per_token": 0.00000014,
        "output_cost_per_token": 0.00000028,
        "litellm_provider": "openai"
    }
}

Adjust the values based on your model. The more info can be found in Aider's docs.

Adding as much configuration as possible directly into Model Library is the main idea to have it there, so I will try and incorporate this into the next release too.

EDIT: Actually, although the Max Output Tokens can be currently set in the Model Library, they are not applied properly when using Aider, so that's something I need to fix.

[scameronde]

Thank's alot for your help and for your fantastic work.
I am looking forward to your "fix" (I would call that a feature).