View Source AI.Agent.Planner (fnord v0.6.4)
The Planner Agent is a multi-modal agent that assists the Coordinating Agent in managing the research process.
All responses from this agent are prefixed with this preamble:
"From the Planner Agent: "
Modes
Prompt (:prompt
)
The Planner Agent analyzes the user's question or prompt, breaks it down into logical parts, and expands it into a list of logical questions that must be answered to provide a complete response.
Initial (:initial
)
The Planner Agent selects and adapts an initial research strategy to guide the Coordinating Agent in its research process.
Check-in (:checkin
)
The Planner Agent evaluates the current research, refines the research strategy, and instructs the Coordinating Agent on the next steps in the research process, including identifying when the research should be considered complete and the Coordinating Agent should respond to the user.
Evaluate (:evaluate
)
The Planner Agent double-checks the validity of the Coordinating Agent's response to ensure that everything it claims is factually accurate.
In this mode, the response
will be either the string "VERIFIED"
or an
explanation of how the response was not factual.
Due to the realities of requiring an AI to respond with valid JSON, the
Planner Agent will make up to 3 attempts to verify the response. If the
response is not in the correct format after 3 attempts, get_response
will
return the normal response map and an appropriate error message.
A "false" response indicates that the Coordinating Agent's response was not factual and should be discarded.
Finish (:finish
)
The Planner Agent saves all relevant insights and findings for future use. In this step, the response member of the return map should be ignored, as it is not intended for the user or the Coordinating Agent.