Release v0.37.0

[wladimiiir]

Highlights

✨ Introducing Task Management: Revolutionize Your Workflow!

We're thrilled to introduce a brand-new Task Management system, replacing the previous session functionality with a more integrated and powerful approach. Now, each of your projects can effortlessly keep track of an unlimited number of tasks (formerly sessions). This allows you to:

• Organize All Your Work: Easily archive previous work or pick up exactly where you left off on any task.
• Boost Productivity: Run multiple tasks simultaneously! Let the Agent tackle one task in the background while you plan or work on another. This dramatically increases your efficiency and helps you get more done.

This is just the beginning for Task Management; we have ambitious plans to further enhance this feature, including individual model and mode settings per task. Stay tuned for more exciting developments!

🧠 Auto-Compacting Conversation History

Say goodbye to overflowing context windows! AiderDesk now supports auto-compacting of your conversation history when token usage exceeds a configurable threshold. This intelligent feature automatically summarizes your long conversations during Agent runs, ensuring the AI always has the most relevant context without being overwhelmed. You can set your preferred token threshold or disable it if you prefer manual context management. Thanks to @pcfreak30 for the suggestion.

Other Changes

• Improved API key handling from environment variables in Agent mode. Thanks to @ohnum for getting this done.
• Fixed an issue where the weak model prefix was not being set correctly when different from the main model.
• Added the possibility to lock the weak model, providing more control over model selection. Thanks to @MotherSoraka for the suggestion.

💡 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 0.37.0.

[fredkzk]

The auto-compacting feature is great! How does the summarization happen behind the scene? Bullet points? Does it follow some aider-desk default instructions? Could we configure it ourselves down the road, with like more or less "compacting ratio" for exampel?

[wladimiiir]

@fredkzk
There is some information about it in the docs. The compact has been around for a while, but in this release you can also make the agent to do it automatically, when the threshold is reached. The system prompt for the compaction can be seen here.

Basically, it will try to get the most relevant information about the current progress, so it can continue further. In the result, it will keep only your user prompt and the compacted information about its progress.