v2.2.0

Features

• Diligence System (Experimental)
  • The llm service class has a new sub service called diligence.
  • You can activate different diligence systems through Bot().llm.diligence.
    • Note diligence can have both positive and negative effects and should be used with caution
    • Example Bot.llm.diligence.stop_word_removal.activate() will activate the stop word removal in the llm connector.
    • Bot.llm.diligence.vowel_removal & Bot.llm.diligence.second_pass

v2.1.0

Features

• CLI !!!
  • Use dandy to access the new command line interface.
  • All commands have a basic help provided with autocomplete.
  • The default / no command will answer questions about the cli.
• Diligence System (Experimental)
  • The Bot class has a new attribute called diligence which is defaulted to 1.0
  • You can adjust diligence between 0.0 (almost no effort) and 2.0 (maximum effort)
  • Adjusting the diligence level allows you to control processing with any LLM.
  • This feature is experimental and works with any llm model.

Fixes

• Updated the dandy.conf.settings to be much more flexible with different use cases and systems.
• Lots of refactoring, typing and code cleaning to new ruff and ty configuration.

v2.0.0

Major Release

• This is a major release, and we highly recommend reading our documentation before upgrading.

Breaking

• Ollama API support has been removed from this project and is now defaulted to the OpenAI API standard.
• Since Ollama supports the OpenAI API standard, you can continue to use Dandy with Ollama.
• Removed calculator module.
• Removed processor module and BaseProcessor class; Bot is now the primary entry point.
• Adapted Decoder from standalone processor into a service usable via the Bot module.
  • Bot().llm.decoder
• Removed Agent module.
• Removed all LLM_DEFAULT_* from settings and now require it to be set inside of LLM_CONFIGS for each model.
  • By default, it uses the defaults on the llm endpoint.
• All exceptions that were postfixed Exception are now postfixed Error.
  • Example: DandyCriticalException is now DandyCriticalError
• The example project has moved to tests.example_project and has been added as a required test.
• Removed PromptOrStr and PromptOrStrOrNone TypeAlias's.
• Removed toolbox module (functionality replaced by the new CLI).
• Removed makefile.
• All Prompt methods have had the argument triple_quote changed to triple_backtick.
• All attributes on Bot that were prefixed with llm_ have had their prefix removed (except llm_config).
  • Example: Bot().llm_task is now Bot().task

Changes

• All options in LLM_CONFIGS now need to be inside an OPTIONS key and are set as lower case keys.
  • Example: OPTIONS: {'temperature': 1.4, 'top_p': 0.7, 'frequency_penalty': 0.2, 'presence_penalty': 0.1}.

Features

• FileService is now available on Bot via Bot().file to make it easy to manipulate and work with files.
• HttpService is now available on Bot via Bot().http for making HTTP requests.
• IntelService is now available on Bot via Bot().intel for manipulating Intel classes and objects.
• The Bot().llm.prompt_to_intel method now supports Vision and Audio.
• The command line interface is back and better than ever, check it out by typing dandy in your terminal.
  • Type into the input to learn more about the features.
  • Use / to run a command.
• BaseListIntel is a new base class for creating Intel objects that primarily wrap a list of items, providing list-like access.
• DefaultIntel is a simple Intel class with a text field for quick use.
• The BaseIntel has some new convenience methods save_to_file and create_from_file for easy long-term storage.
• Configuring LLM options can now be done through Bot().llm.options and Bot().llm.decoder.options.
  • Example: new_bot = Bot() ... new_bot.llm.options.temperature=1.4

Fixes

• Fixed a bug with directory_list method on the Prompt class when a file has no extension.
• dandy.llm.conf.LlmConfigs are now checked during usage to allow for better control loading environments.
• dandy.conf.settings now reloads its current state at the time of attribute access instead of once during init.
• Fixed many issues with customizing all Service and Processor subclasses.
• Decouple a lot of the LLM modules to allow for better maintainability and testing by using the new LlmConnector.
• Added 2026_roadmap.md to track future development plans.
• Reorganized project structure to be more service-oriented.

v2.0.0.alpha.0

Major Release

• This is a major release, and we highly recommend reading our documentation before upgrading.

Breaking

• Ollama API support has been removed from this project and is now defaulted to the OpenAI API standard.
• Since Ollama supports the OpenAI API standard, you can continue to use Dandy with Ollama.
• Removed calculator module.
• Removed processor module and BaseProcessor class; Bot is now the primary entry point.
• Adapted Decoder from standalone processor into a service usable via the Bot module.
  • Bot().llm.decoder
• Removed Agent module.
• Removed all LLM_DEFAULT_* from settings and now require it to be set inside of LLM_CONFIGS for each model.
  • By default, it uses the defaults on the llm endpoint.
• All exceptions that were postfixed Exception are now postfixed Error.
  • Example: DandyCriticalException is now DandyCriticalError
• The example project has been removed.
• Removed PromptOrStr and PromptOrStrOrNone TypeAlias's.
• Removed toolbox module (functionality replaced by the new CLI).
• Removed makefile.
• All Prompt methods have had the argument triple_quote changed to triple_backtick.

Changes

• All options in LLM_CONFIGS now need to be inside an OPTIONS key and are set as lower case keys.
  • Example: OPTIONS: {'temperature': 1.4, 'top_p': 0.7, 'frequency_penalty': 0.2, 'presence_penalty': 0.1}.

Features

• FileService is now available on Bot via Bot().file to make it easy to manipulate and work with files.
• HttpService is now available on Bot via Bot().http for making HTTP requests.
• IntelService is now available on Bot via Bot().intel for manipulating Intel classes and objects.
• The Bot().llm.prompt_to_intel method now supports Vision and Audio.
• The command line interface is back and better than ever, check it out by typing dandy in your terminal.
  • Type into the input to learn more about the features.
  • Use / to run a command.
• BaseListIntel is a new base class for creating Intel objects that primarily wrap a list of items, providing list-like access.
• DefaultIntel is a simple Intel class with a text field for quick use.
• The BaseIntel has some new convenience methods save_to_file and create_from_file for easy long-term storage.
• Configuring LLM options can now be done through Bot().llm.options and Bot().llm.decoder.options.
  • Example: new_bot = Bot() ... new_bot.llm.options.temperature=1.4

Fixes

• Fixed a bug with directory_list method on the Prompt class when a file has no extension.
• dandy.llm.conf.LlmConfigs are now checked during usage to allow for better control loading environments.
• dandy.conf.settings now reloads its current state at the time of attribute access instead of once during init.
• Fixed many issues with customizing all Service and Processor subclasses.
• Decouple a lot of the LLM modules to allow for better maintainability and testing by using the new LlmConnector.
• Added 2026_roadmap.md to track future development plans.
• Reorganized project structure to be more service-oriented.

v1.3.5

Fixes

• Fixed the Decoder prompt and intel classes to properly match each other.

v1.3.4

Fixes

• Cleaned up the default llm_role and llm_task instructions to be more generic and less intrusive.
• Cleaned up the Decoder system prompt to be more accurate.
• Cleaned the main llm service system prompt to improve results across Dandy.
• Added the missing line break when adding a Prompt to a Prompt for better formatting.
• Changed the formatting on the Decoder prompt to produce better results when there is no max_return_values
• Fixed a bug with decoder not rendering recorder events properly.
• Corrected recorder failing to render OpenaiRequestMessage structure.
• Fixed the problem with setting custom attributes on instantiating processors.

v1.3.3

Fixes

• Fixed bug causing system_override_prompt to not be handled properly in the llm service.

v1.3.2

Fixes

• Fixed the system_prompt_override to properly override the system prompt.
• Fixed the Decoder prompt to properly show plural when there is no key limit.

v1.3.1

Fixes

• Fixed and order of operations mistake with generating the llm service system prompt.

v1.3.0

Breaking

• Service method llm.prompt_to_intel no longer takes postfix_system_prompt as an argument as its redundant.

Features

• All Processor classes Bot, Agent and Decoder now have a reset_services method.
  • This provides an easy and lightweight way to reset any persistent data in all the processor services.
• All Service classes Llm, Http, Intel now have a reset_service method.
  • This is for having more fine control over persistent data.
  • there are other more fine control reset methods, and we recommend you check out the API reference.

Fixes

• Fixed the signatures for __init__ method on Agent, Bot and Decoder.
• Correct the order of operation for when message history adds responses in the llm service.