Dandy

Features

- Added `object_source` method to `Prompt` to get the source of a specific object in a module.
- Drastically improved the LLM service to retry and handle prompt correction.

Fixes

- Fixed the `dandy.utils` to correct a circular import problem.
- Fixed a problem with the `file` method on the `Prompt` class to search the `settings.BASE_PATH` directory.

Breaking

- The `debug` module is now the `recorder` module.
- This also includes many refactors like `DebugRecorder` to `Recorder` ETC.
- In settings `ALLOW_DEBUG_RECORDING` has been removed.
- In settings `ALLOW_RECORDING_TO_FILE` hase been added with a default value of `False`.
- The `destroy` method on `Cache` object has now been replaced with `destroy_all` for clearing all caches.

Features

- `DebugRecorder` is now `Recorder` and has new rendering methods and overall improved rendering quality.
- You can now render to json, html and markdown files and strings for easy exporting.

Changes

- `SqliteCache` now uses one file to store all different caches.
- this allows for more fine control in the same way the `MemoryCache` works.

Fixes

- Fixed the llm service to properly handle retrying when an invalid response is provided with progressive exceptions.
- Improved the `BaseIntel` class to have much better handling for complex operations.

Breaking

- Moved 'dandy.core.cache' to 'dandy.cache'

Fixes

- Debug recorder correctly renders OpenAI service request messages.
- Added more testing for more complex JSON Schemas.
- Fixed a bug where the `Prompt.prompt` method would not accept strings properly.

Fixes

- Fixed problem with `BaseIntel` required fields using nested includes and excludes.

Fixes

- Fixed bug with using include and exclude on intel objects that did not validate filled and empty fields properly.

Breaking

- Removed the `assistant_str_prompt_to_str` method and sub methods from `LlmService`.
- Everything should be processed through the `LlmBot` going forward.

Features

- Added a new processor called `LlmMap` that uses the `Map` object to easily create contextual relationships to python objects.
- Debug recorder now shows the JSON schema that was sent to the llm service.
- New decorator called `debug_recorder_to_html(debug_name: str)` that allows you to easily wrap a function or method.

Changes

- Removed the Contrib Selection llm bot.

Fixes

- Fixed a caching to be more robust and understand changes with in dandy processors for more accurate caching.
- Drastically improved the testing of the OpenAI llm service.