Some NeurContainers contain clear entrypoints, while others point to general directories of binaries which can be executed.
For the LLM to reliably call tools it need to know the expected description of the general function and the arguments.
Since it's currently not feasable to handwrite or review this for all tools we would need a (semi-)automated way to add tools to an index that the LLM can use for tools discovery.
This is an open ended issue and a final implementation will probably require multiple iterations and manual testing.
Some NeurContainers contain clear entrypoints, while others point to general directories of binaries which can be executed.
For the LLM to reliably call tools it need to know the expected description of the general function and the arguments.
Since it's currently not feasable to handwrite or review this for all tools we would need a (semi-)automated way to add tools to an index that the LLM can use for tools discovery.
This is an open ended issue and a final implementation will probably require multiple iterations and manual testing.