v0.2.2

Changes

• Update requirements.txt to be more specific, resulting in correct dependency resolution without manual intervention.

v0.2.1

Changes

• Updated dependencies and confirmed that project still works as expected.
• Reduced default multimodal token limit to 1000 to encourage LLM to focus on the current image.

v0.2.0

Features

• Vision/Multimodality: Neuro will now take screenshots and send it to a multimodal llm for processing! Great for reaction content or live commentary.

Changes

• Refactored llmWrappers. TextLLMWrapper and ImageLLMWrapper are subclasses of AbstractLLMWrapper. This should allow for easier extensions onto other LLMs with differing API formats.
• Updated the RealtimeTTS and RealtimeSTT libraries, which solves some dependency headaches. If you have dependency troubles upgrading from an existing installation, trying removing all packages and reinstalling following the installation instructions.

Note: Release versions do indicate that it has been end-to-end tested, but I still recommend cloning main for bug fixes and tweaks that make the AI behave better.

v.0.1.0

Features

• Memory / RAG: Persistent memories are automatically created and can be manually added for recall by the AI, powered by ChromaDB.

Fixes

• Blacklist filtering filters substrings (If "war" was blacklisted "warmth" would be filtered).

I will be creating minor version releases for major feature additions/enhancements moving forward. Since this project is still v0.x.x, there are no guarantees of backwards/forwards compatibility.