v0.3.6

chrome extension: add dynamically change context processing params for difference page content types currently support only text/html \n\n

server: update html agent logic, change prompts, add self reflection via structured outputs(checklist) and streaming that type of response, major changes in html processing, add processing as text in html processor

Full Changelog: v0.3.4...v0.3.6

v0.3.2

Full Changelog: v0.2.6...v0.3.2

v0.2.6: Chrome extension update:

- added inspection of html page elements, now you can load into the context not the entire page but a specific element, also when changing the tab it is not reset
- removed the dependency in html processing on the head and body tags

v0.2.4 First good release for long pages, and finally full page processing

changed my views, it is hard to find normal embedding models for my purposes, also with simple html processing in md, a lot of information is lost, so I split raw html into chunks not exceeding the maximum size value, so we really consider the html page and not just its main text content and we can directly ask questions about page elements, however there are problems with prompts and with the behavior of the model and with preprocessing I tried to use attributes but for now I will just forget about it if anything you can comment the lines in parse_tag in htmlprocessing. In future releases I plan to add several features to the extension, such as selecting a specific html element as in the developer tools in the browser and selecting multiple pages for context. On the server side, you still need to think about how to preprocess html and in what form to submit it all to the model, perhaps replace LLM with another, and make chat_complete to automatically format the text for submission to llm