v2.2.1 Release

This release aims to fix many of the issues found in v2.2.0.

Better prompting

While not encompassing all changes planned, this aims to resolve the discrepancy in expected prompt inputs for various APIs. Prompter now only handles an instruction prompt and the conversation history. Conversation history is not converted into text until inside T2T-based operations to allow per-implementation handling of history. This allows for interchange between instruct and chat based models and improves model adherence to intended response format with OpenAI models.

Fixed operations/use endpoint

This endpoint was not fixed during the refactor and was thus out of spec. This was fixed and updated given the new T2T operation spec.

Fixed installation and examples

Installation should now work cross platform. Examples and development documentation have all been corrected.

v2.2.0 Release

Introducing MCP support!

Any MCP server can now be added to jaison-core through a few simple steps. An MCP client has been integrated for tool-calling prior to response generation. All server tools and resources usage and descriptions are automatically injected into the prompts, so all that needs to be done is to download the server and configure the YAML.

Additional features have been added such as support for MeloTTS, OpenAI embedding models, and support for per-operation configuration.

Bug fixes and changes:

• Downgraded to Python 3.10
• Updated installation procedure
• Fixed issues with broken dependency builds
• Fixed faulty history file encoding
• Added new websocket events in response pipeline for MCP information

v2.1.0 Release

A complete and polished rewrite!

We've decided to discard the changes introduced in v2.0.0 and opted to polish the prototype, keeping the simple and maintainable design, but restructuring so its extensible. This repository now only contains server logic, and front-facing applications have been moved to their own repositories. These include the features you already know, and have extended them with more integrations and lowered latency. Since this is a complete rewrite, it would be too much to go through all the changes here, but read the many helpful documentation in this repository to familiarize yourself with what's included!

v2.0.0 Release

Introducing Plugins!

Project J.A.I.son is now plugin driven. Backend models responsible for generating responses have been moved to component plugin projects, and front-end reactives have been moved to application plugin projects. Please refer to the new README for installation guide.

v1.6 Archive

This is an archive of v1.6 following some bug fixes. This version is the final version of the prototype encompassing all features created over versions 1.X.

v1.5 Archive

This is an archive of v1.5. This is a very unpolished version of v1.6. This version adds a web UI that allows configuration and control over the system. Features include green-screenable subtitles, audio-response multiplexing, response preview, feature toggles, configuration editor, prompt editors, and more.

v1.4 Archive

This is an archive of v1.4. This version adds integration with VTube Studio. Responses from the AI are converted into animations depending on the emotion, automatically cycling a VTuber model through animations based on the responses.

v1.3 Archive

This is an archive of v1.3. This version adds speech-to-text, automatically taken speech input from users in a voice call, adding to memory, and automatically requesting a response. The Discord bot was refactored to use discord.py instead of pycord.

v1.2 Archive

This is an archive of v1.2. This version adds text-to-speech, allowing users to hear responses spoken into a voice call. This supports OpenAI and pytts generated speech, and includes a client interfacing with the RVC Project for voice conversion.

v1.1 Archive

This is an archive of v1.1. This version adds better support for T2T OpenAI backends and Unsloth backends. This also contains some example scripts for creating datasets.