roadmap.todo

V0.3.1
[ ] - fix memory workflow

[ ] - pygmalion grammar support

[ ] - togetherai model


V0.4:

[ ] - outlines as (optional) backend for guided generation (mostly json-schemas)

[ ] - dspy integration for autotuning prompts

[ ] - pydantic model streaming  (6h)

[ ] - enable union type without function calling  (6h)

[ ] - enable Error type for non union calls  (4h)

[ ] - convert langchain tools to funcchain agent/router  (8h)


V0.5+:

[ ] - cookbooks folder with jupyter notebook tutorials  (8h)

[ ] - depends functionality to create nested chains and compile into runnables  (10h)
        # add a deps thing to put into funcchain defs that takes another chain and compiles it into a runnable
        # so langsmith shows nested chains
        # in the chain creation process it just runns all of the depending chains in parallel and feeds the inputs into sub chain
        # output of every chain is string when inserted

[ ] - vector retrieval type to add file/url context into chains  (20h)

[ ] - implement variable compression to inject as much context as possible without overloading  (20h)
        # Create a Compressable Context Schema where it is compressed in case the context lenght is already filled
        # So anything that is additional can be compressed to fit in the context but when other things that are important are not compressed.
        # Optionally you can define how to compress and where to leave the gaps (default in the middle with [...])

[ ] - LLMCompiler written in funcchain example  (30h)

[ ] - vscode extension for custom syntax highlighting  (30h)

[ ] - parallel function calling  (8h)

[ ] - FuncUnion and str output  (6h)

[ ] - think of new syntax for special dspy modules (COT, FewShot, ...)