Release v0.40.0

[wladimiiir]

Highlights

🌡️ Enhanced Model Configuration: Temperature & Max Tokens at Your Fingertips

Gain unprecedented control over your AI's behavior! You can now configure temperature directly for each model in the Model Library, allowing for precise fine-tuning of response creativity. We've even added default optimal values for specific models like Gemini. Plus, the Max Tokens setting in agent profiles is now optional, defaulting to the model's configured value. This streamlines your workflow, as you no longer need to adjust these settings every time you switch models.

🚀 Deeper Aider Integration with Model Library Settings

We've significantly improved how Aider Desk's Model Library integrates with the Aider process. Aider now correctly utilizes the costs and temperature settings directly from your Model Library, reducing the need for separate Aider-specific model configuration files. Our goal is to make the Model Library your single source of truth for all model parameters. If there are other parameters you'd like to see configurable here, please let us know!

✨ Streamlined New Task Creation

Creating new tasks is now more intuitive and efficient! When you initiate a new task, it immediately appears as [Untitled] in your task sidebar, allowing you to easily save it by simply setting a name. This also brings a significant advantage for Git Worktree users, as the new branch will be created based on the task name, providing a clearer and more organized project structure.

🤝 New Provider Support: MiniMax

We're excited to announce direct support for the MiniMax provider! MiniMax offers a "Coding Plan" subscription, making it a compelling option for those seeking cost-efficient yet capable model. Our implementation utilizes Anthropic's compatible API, ensuring a smooth integration. Definitely worth checking out for your coding tasks!

Other Changes

• Added maxResults parameter to the grep tool, with a default of 50, for more controlled search results.
• The "Show/Hide all files" state now persists across tasks within the same project.
• Improved utilization of abort signals and enhanced interruption handling for a more robust experience.

💡 We’d love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

---

This discussion was created from the release 0.40.0.

[scameronde]

I tried the update and browsed through the code changed (without understanding enough), but I do not see that "max_tokens" - as manually defined by me in the model library - is send to the coding model of aider.

Am I right with this, or do I miss something?

And again, thank your for continually improving this awesome tool.

[wladimiiir]

You are right. As mentioned in the release notes only costs and temperature is passed to Aider. The changes were bloating when I was implementing everything related around that, so I had to stop at one point to be able to finish the initial part.

That said, I will include max tokens as another parameter passed to Aider. Let me know, if you see any other parameters missing. Thank you.

[scameronde]

I really thought about that, but "max_tokens" is the only one that gives me big problems. The only other parameter I can think about is the 'alias' "max_completion_tokens", but that is basically just the same and would be there to be future proof.

[wladimiiir]

@scameronde
After checking the code, it should actually be setting also the max_tokens. I was doing some additional work for the temperature, that's what confused me and I thought I didn't set this parameter.

But in order to make it work, you must not have any .aider.model.metadata.json or any other model config file present with the model; otherwise AiderDesk will not overwrite the model parameters (this is by design). Let me know if that's not your case and this is still not working for you and how you are testing it. Thanks.

[scameronde]

Sorry for answering that late.

1. I checked if I have any aider model configs in my project or home dir (did a find over everything). Deleted all.
2. restarted AiderDesk
3. switched into Ask mode
4. asked it the show me something that generates a lot of tokens

got (consistently)

Model openai/aider-claude-haiku-45 has hit a token limit!
Token counts below are approximate.

Input tokens: ~42,085 of 200,000
Output tokens: ~3,312 of 63,999
Total tokens: ~45,397 of 200,000

What is my setup:

• I am using the "openai-compatible" connector to connect to our companys model proxy (litellm based)
• the model 'openai/aider-claude-haiku-45' points to a Haiku-4.5 instance hosted by Google in Europe

I talked with the team that provides the proxy. They say

• they don't mess with the max_tokens parameter
• they don't block the max_tokens parameter
• Google Europe seems to be the one limiting the max_tokens

I then did some tests with a local Open WebUI instance and wrote some code accessing this model using our proxy:

• when not setting ANY parameter, I also hit the 3xxx token wall
• when setting max_tokens in Open WebUI or in my code, I am getting the max amounts of tokens

Do you have any idea what I am doing wrong?

[scameronde]

One add on: the max token limit of 63999 was set by me deliberately, just to check if it used at all