ml-commons
ml-commons copied to clipboard
[FEATURE] Flow agent Enhancement
Is your feature request related to a problem?
Background
A flow agent is configured with a set of tools that it runs in order. The agent coordinates the tools so that one tool’s output can become another tool’s input, via putting the output in the params
map.
Problem
However, its implementation is not robust and lacks some function currently. For example:
- No input validate check
- It's not secure that one tool can access all output from the tools previous to it but not just the output of the immediately preceding tool.
- The execute
params
map will get larger and larger if the flow becomes longer - If having a MLModelTool in this workflow,
- it should have the description of both previous tool and next tool in its prompt context automatically.
- it should know where to obtain the output from the previous tool and put that in its prompt
- it should know its output format(i.e, the input requirements of the next tool) and put that in its prompt
What solution would you like? TBD. A clear and concise description of what you want to happen.
What alternatives have you considered? TBD. A clear and concise description of any alternative solutions or features you've considered.
Do you have any additional context? Add any other context or screenshots about the feature request here.