## Notes - there was a mechanism that combine chunks together after summarization and if combined chunks were larger than token budget, summarizer tried to split to new chunks and summarize again. This didn't work as expected and caused an infinite loop - I removed this crazy logic and also I added preprocessors that clean text before it is sent to LLM. - the next bug was related to broken pipe. It was not possible to pass output from previous process to nitrodigest using pipe. I fixed it and changed CLI API and from now it accept input as an argument instead of flag. - The next thing I fixed was related to broken progress indicator during processing multiple emails. Even if there was an error and email was not processed, nitrodigest printed that all emails were processed successfuly. - I wrote a bash script that simplifies reading emails using himalaya. It's published as gist. Documentation is updated - At this moment I consider Nitrodigest first version as done. - The next step is to promote it and find users that will want to use it and give some feedback ## Related pull requests - <https://github.com/Frodigo/garage/pull/177> - <https://github.com/Frodigo/garage/pull/178> - <https://github.com/Frodigo/garage/pull/181> ## Related issues - <https://github.com/Frodigo/garage/issues/166> - <https://github.com/Frodigo/garage/issues/161> - <https://github.com/Frodigo/garage/issues/157> - <https://github.com/Frodigo/garage/issues/167>