v0.7 is ready for testing in testing branch on github. Main branch remains unchanged. I couldn't make a video today, so I'll try tomorrow, here's a quick list of the changes:
- UI improvements: biggest changes are design revamp by and utility framework messages that you can turn on/off. Some bug fixes, automatic autoscroll switching based on scroll position in window. Also a progress indicator is added right above the chat input to show you what is going on right now. - automatic memory: agent can now remember memories and solutions after each monologue and recall them at the start of each monologue. It will also replace older memories that are very similar by newer versions so it does not grow unnecessarily.
- memory management: agent has access to all areas of memory including solutions and can manage them. Thresholds have been fixed, memory areas, filter conditions, timestamps added, memory formatting improved.
- knowledge import: files inside knowledge folder are automatically imported to agent memory. Knowledge folder now has subfolders for default and custom knowledge files, default is for framework pre-built knowledge, custom is for you and is ignored by git. Under these folders are subfolders by memory areas, for example solutions. This way you/I can tech A0 lots of solutions to problems and it does not pollute the system prompt as they are only recalled when needed. I added example file for youtube download procedure.
- extension framework: /python/extensions folder now contains subfolders where scripts can be placed and these will be executed at various positions of the message loop process. This way the main agent.py code will stay clean and short and all the memory extensions and features can be developed independently. - tool prompt split into one file for each tool, all files named agent.system.tool.*.py are loaded automatically
A lot has changed and needs to tested before merging to the main branch. If you find any time to try and report bugs, I'll be grateful:)