V0.3 Agentic & Source-Aware Research

Research gets deeper and more controllable, with multi-step agents, explicit source selection, dataset-aware analysis and smarter model routing.

Agent-assisted research – Added multi-step research capabilities for questions that require more than a single model response.
Dataset-aware research – Lucid can use selected datasets as part of a research task rather than treating every question as generic web research.
Source controls – Added clearer ways to define which available sources and connected services should contribute to an analysis.
Expanded model routing – Improved how Lucid selects and works with different models depending on the type of research being performed.

More deliberate research workflows – Improved the distinction between direct answers, deeper research and work requiring multiple steps.
Better source architecture – Strengthened how selected datasets, web research and connected services flow through the Assistant.
Clearer Signals setup – Refined how monitoring prompts communicate what to track, what should matter and which information sources are relevant.
Improved entity understanding – Continued strengthening recognition of people, organisations and other entities across research.

Fixed model-routing issues caused by retired or unavailable model versions.
Improved fallback behaviour when a preferred research path could not be used.
Fixed inconsistencies between selected sources and sources available to downstream research.
Strengthened reliability around long-running agent-assisted requests.