Dustin Rush
Dustin Rush
@jxnl I think this is ready.
I need this as well. Will look at implementing something.
+1 - v2 looks very promising indeed
related I just tried out the [MCP inspector](https://github.com/modelcontextprotocol/inspector) last night and it worked really well for testing.
If someone does pick this up, it would be nice if there was a way to only convert some endpoints to tools. Also if the tool could accept and return...
Love it! State, nodes and edges is what we need. Was looking at pydantic-ai's [pydantic_graph](https://github.com/pydantic/pydantic-ai/tree/main/pydantic_graph) sub library for this, but haven't implemented anything from it yet. Happy to talk more...
There are two types of agents we're building: 1. Agents that have tools and decide which one to call for a period of time (even infinitely) 2. Workflows which are...
For further context, this is how we are handling different workflow processes in parallel (like processing multiple documents at once): https://docs.hatchet.run/home/features/child-workflows#scattergather-workflows One problem is that paths have to be predetermined...
Here are some more thoughts: - errors: how would you handle errors? fallback nodes? - being able to define callbacks for agent (node) completion or errors has been useful for...
One thing about defining the handoff there may be some context grooming for the handoff (e.g. premium subscriber vs basic subscriber may have different workflows with different context needs). So...