kobold_assistant

Full Changelog: https://github.com/lee-b/kobold_assistant/compare/v0.1.3...v0.1.4

• Add sleep mode, and sleep/wake commands
• Fix issue with hallucination mitigations
• Catch more hallucinations
• Add an assistant description separate from the main prompt
• Improve handling of Ctrl-C
• Improve assistant laughter
• Fix default log level
• Lower default assistant temperature ("craziness") for more factual responses (by default, that is, configure as preferred)
• Greatly clean up the terminal output, to basically just the dialog
• Colorize terminal output
• Expose TTS speech speed as a configurable setting
• Decrease assistant speech speed
• Update dependencies
• Properly use python logging, except for core program output to user
• Handle thinking emoji returned by some language models
• Don't prompting the user that assistant is still there during sleep
• Drop the full determinism metaparam that was commented out
• Stop augmenting settings internally
• Document various errors and what they mean
• Document known-bad LLMs
• Remove trailing blank lines in settings
• Improve prompt for instructional models
• Reduce default temperature, for more sanity.
• Send stop words and other metaparams to the llm
• Catch Whisper exceptions, report them cleanly, and retry
• Add the concept of "souls" - unique assistant seeds, that can be stored in settings (WIP, do not use yet)
• Add sleep mode and sleep/wake commands
• Mention the KoboldAI discord server channel in docs
• Add max length settings
• Various other small improvements and fixes
• Pronounce real numbers and various other small improvements
• Document some common known-good models and how to use them.
• Update README.md
• Mention the KoboldAI discord server channel

Add a proper configuration system using json files and configuration files in /etc/, the user's home directory, with priority-/locality-based override order, and dependency resolution of template variables used in the configs (for user and assistant names, etc.).

Support KoboldAI's API, not just KoboldCPP. Improve mic handling.

• Support the KoboldAI API, not just KoboldCPP.
• Autodetect the first suitable microphone, by default.
• Support auto-calibrating the microphone, and do that by default.
• Add a setting for SLOW_AI_RESPONSES, off by default, for quicker models / machines where realtime conversations are possible.
• Improve the prompt to try to tell the AI how to joke around.
• Plus various other small changes

• Add selection of microphone devices via configuration
• Add a list-mics command line mode, for finding out which devices can be configured
• Various documentation improvements

First release