Roadmap

The Yacana adventure is merely just starting.

We expect to release the following new features in the next update:

• Compatibility with vision models like llama3.2-vision. [Status]: Done. Available on dev branch.
• Compatibility with VLLM [Status]: In progress. We need to shift from the Ollama client library to allow the OpenAi one.
• Rework JSON mode to allow "structured output". [Status]: Todo. This will allow better tool calling success. In particular with VLLM which does not support the JSON mode.
• Compatibility with OpenAI. [Status]: Todo. It should be simple to develop after adding VLLM support as it must be in "OpenAI mode".
• Monitoring Yacana workflows to track down prompt engineering failures like tool calling failures count, etc. [Status]: Todo. Low priority.
• Simplifying shift message system and maybe reworking GroupChat itself a bit. [Status]: Todo.
• Keep working on the documentation.
• Add a section on code generation. In theory, we won't add any abstraction because this can be achieved using the current features but still needs documentation. FYI just call a tool that takes a string in input. The string would be the code to execute. Execute the code inside the tool. Problem solved... [Status]: Todo.
• More to come

Related to Yacana:

• Updating Gno6 as it's a good demonstration of what cool stuff can be achieved with Yacana
• Making a RAG tool using Yacana
• Making another smart CLI. Either GIT or Docker maybe.
• More youtube videos to come
• More to come