Research Architecture (Multi-agent+Multi-graph)

[TanGentleman]

Loosely defined:

• Tools: The bridge between LLM output -> An action, like read/writes to a database. Any function can be a tool.
• Agent: Executes a workflow given a task, restricted by a fixed set of tools.
• Graph: A set of defined relations between defined nodes. The structure and content of the node depends on the use case.

Food for thought. In my opinion, each of these excel in their intended space:

Complex retrieval tasks over a Knowledge Graph/Table - WhyHow Organized workflow automation - n8n Orchestrating agents with good cookbooks - composio

[TanGentleman]

Here's the design I'll be conforming to for the first agentic endpoint in the works: https://agentprotocol.ai/sdks/python

[TanGentleman]

Managing state is hard. Here's one approach: https://langchain-ai.github.io/langgraph/how-tos/human_in_the_loop/time-travel/