A finished report is a starting point, not an endpoint. LumaVista’s project chat panel lets you have a conversation with your research — asking follow-up questions, requesting changes, and exploring findings in more depth. The chat has full context of everything that happened during your project: the original prompt, every search result, every reasoning step, and the final report.

Opening the chat panel

The chat panel is accessible from any project view. For running projects, you can chat while the research is still in progress. For completed projects, the chat picks up right where the research left off.

The panel appears on the right side of the project view on desktop, or as a tab on mobile devices. Type your message in the input area at the bottom and press Enter or click Send.

What the chat knows

The chat is not a generic AI conversation. It is grounded in your specific research project. When you ask a question, the chat draws from:

Your original prompt — The structured research question, including sections, exclusions, and scope.
All agent outputs — Every search result, reasoning analysis, and synthesis produced by the agents during the project. This includes results from nodes at every level of the research graph.
The final report — The structured report produced by the report writer agent.
The research graph structure — The chat understands which agents produced which findings, how subtopics were organized, and what areas were explored versus skipped.

This means the chat can answer questions about things that are not in the report itself — for example, search results that were found but not included in the final synthesis, or reasoning steps that informed a conclusion without being explicitly stated.

Types of follow-up questions

Here are the most useful ways to interact with the chat:

Clarifying findings

Ask about specific claims or sections in the report:

“What sources support the claim about market growth in section 2?”
“Were there any contradicting sources on the cost comparison?”
“What is the confidence level on the regulatory timeline?”

The chat traces back through the agent outputs to find exactly where a finding came from, which sources informed it, and whether there were dissenting views.

Requesting deeper exploration

If a section feels thin, ask for more:

“Can you expand on the section about supply chain risks?”
“I need more detail about the European regulatory landscape.”
“What else did the search agents find about competitor pricing that did not make it into the report?”

The chat can pull in findings from the research graph that were not included in the final report — search results that were deemed secondary, or reasoning that was summarized rather than included in full.

Challenging conclusions

Push back on the report’s analysis:

“The report suggests X, but I have seen evidence of Y. How do you reconcile that?”
“This conclusion seems to rely heavily on a single source. Are there corroborating sources?”
“The competitive analysis ignores company Z. Why?”

The chat explains its reasoning and acknowledges gaps. It is not defensive about the report — if the research missed something, it will say so.

Requesting changes

Ask for modifications to the report content:

“Rewrite the executive summary to be more concise — three paragraphs maximum.”
“Change the tone of section 3 to be more appropriate for a board presentation.”
“Add a comparison table for the three approaches discussed in the analysis.”

The chat produces revised content that you can use directly.

Exploring what was not covered

Sometimes the most interesting questions are about what the research did not find:

“Were there any topics you wanted to investigate but could not due to budget limits?”
“What areas of this topic would benefit from deeper research?”
“Did any of the search agents hit dead ends?”

This can help you decide whether to run a follow-up project focused on gaps in the first one.

Chat context and the research graph

The chat’s context is tied to the project’s research graph. This has a few practical implications:

Larger graphs give richer context. A project that ran Deep Research and expanded through several depth levels gives the chat more material to draw from. A quick Landscape Survey has less context available for follow-up questions.

Rejected expansions limit context. If you used Human-in-the-Loop to reject expansion into a subtopic, the chat’s knowledge of that area is limited to whatever the agents found before the rejection. It will tell you if it does not have enough context to answer a question about an unexplored area.

The chat can go and find more, but it will tell you it is doing so. By default a follow-up question is answered from the research you already have — that is fast and it is grounded. When your question genuinely cannot be answered from the existing context, the chat can search for the gap or commission a fresh research run rather than guessing. Those are visible actions, not silent ones.

Tips for effective chat interactions

Be specific. “Tell me more about section 2” is less useful than “What primary sources support the cost projections in the financial analysis section?” Specificity helps the chat find the right context quickly.

Reference the report structure. If your report has named sections, reference them by name. “In the Competitive Analysis section…” focuses the chat immediately.

Ask about sources. The chat knows which sources informed which findings. Asking “What are the three most important sources for this section?” is a powerful way to evaluate the quality of a finding.

Use the chat to plan follow-up research. If the chat identifies gaps or areas where the research was thin, ask it to help you write a prompt for a follow-up project. It knows exactly where the first project’s coverage fell short.

Limitations

The chat is powerful but bounded:

It does not read your other projects’ research graphs. Its grounded context is this project. It does draw on your knowledge graph, which spans everything you have researched — so it knows what you know, without reading another project’s raw findings.
New research is an action, not a reflex. It will not quietly re-run your project because a question was hard. When it needs more, it says so and does it as a visible step.
Context limits. For very large projects with extensive agent outputs, the chat may summarize older context rather than including it verbatim. It always prioritizes the most relevant context for your question.

Attaching an image to a message

Some questions are easier to show than to describe — a screenshot of a dashboard, a photograph of a whiteboard, a chart from a PDF you cannot copy out of. You can attach an image to a chat message and ask about it.

Availability. Image attachments are in beta and off by default. The paperclip in the composer is always visible; where the capability is not available to you it is shown disabled, with a tooltip saying why, rather than hidden. To ask for it, contact us.

There are three ways to attach, and they all end up in the same place:

The paperclip in the message composer opens a file picker.
Paste a screenshot straight into the message box. The image is attached rather than pasted in as text.
Drag and drop a file onto the composer.

PNG, JPEG, WebP and GIF are accepted, up to 10 MB each and four images per message. Anything else in the same paste or drop is ignored rather than causing an error. You can send images with no text at all, if the question is obvious from the picture.

Attached images are stored encrypted, under your own key, in the same way the rest of your data is — which is also why a locked session cannot accept one: there is no key to seal it with until you unlock.

Which model reads the image is decided by the same routing rules as every other request, so your trust floor governs it exactly as it governs your research. If you need the guarantee that it never leaves EU-sovereign infrastructure, set the floor rather than assuming it — see Where Your Data Is Processed.

Two things will stop an attachment:

A locked session. Your key is what the image is sealed with, so while your session is locked the paperclip is disabled and the tooltip asks you to unlock.
A file of the wrong type or over the size limit. The attachment appears in the composer with the reason it failed; remove it and carry on. Sending waits until every attachment has finished uploading, so nothing is dropped silently.

Once attached, the image is part of that turn’s context. Ask about it the way you would ask about anything else: “what is wrong with this chart?”, “transcribe the handwriting in this photo,” “does this diagram match what section 3 describes?”

Next steps

Once you are comfortable with the chat panel, you can use it as a central workflow:

Run a project with the guided prompt builder.
Review the report and research graph.
Use the chat to clarify, expand, and refine.
Export or copy the refined content for your final deliverable.

For information about exporting and formatting your final report, see Refining Reports.