Back to changelog
Improvement

The agent can stay with harder questions

Multi-step investigations have more room to follow the evidence, recover from a failed lookup, and finish a useful answer.

#Agent#Reliability#Analysis

The grid agent can now go further on questions that need several data lookups or calculations. A failed query no longer cuts the investigation short, and answers continue streaming back as the work progresses.

When a question is too broad to finish in one pass, the agent gives a clearer stopping point so it is easier to narrow the question or continue from the evidence already found.

The live Ask the Grid homepage with a multi-step ERCOT analysis prompt comparing demand, wind generation, and system lambda over 24 hours.
Frame a multi-step question; the agent follows the grid data needed to answer it.