Open maeste opened 2 weeks ago
As mentioned in https://github.com/wise-agents/wise-agents/discussions/222:
Realized that we actually have all of the state related information as part of the context already.
The phases
, max_iterations
, confidence_score
, and optionally the system_message
to use are part of the coordinator agent configuration and aren't related to the state itself. Note that phases only contains the phase names. I'll likely rename this. If a coordinator agent were to go down and be restarted, it would be restarted with these values.
When a coordinator receives a request, it creates a new chat id for the request and adds the following information to the context to keep track of the state. This information from the context gets updated by the coordinator when processing a response.
_agent_phase_assignments
: This maps a chat id to a list that contains a list of agent names to be invoked in each phase_current_phase
: This maps a chat id to the current phase, represented as an index_required_agents_for_current_phase
: This maps a chat id to a list of agent names that need to be executed for the current phase_queries
: This maps a chat id to a list of queries representing the queries attempted in each iteration (this will need modifications as we come up with a strategy for formulating queries specific to each agent)I'll add this information to WiseAgentContext as part of https://github.com/wise-agents/wise-agents/issues/211.
Have updated the title of this issue to be about testing the case where the coordinator agent goes down and needs to be restarted to check that we can handle this case.
Just realized that both coordinators have a route_response_to variable that should be moved to the context. That's included in my PR here: https://github.com/wise-agents/wise-agents/pull/228
This is discussed in discussion #222