Skip to main content

2026.11.14

Product Release Notes

Written by Vurvey Support Agent

New Features and Enhancements

  • Simulations: your customers can run them now

    • The problem: "We want to test our ideas ourselves, without waiting on Vurvey to run a simulation for us."

    • What changed: Before this release, only Vurvey ran Simulations. Now your Vurvey team can turn Simulations on for your workspace, and you can run

  • One Tools button in chat

    • The Images, Sources and Web Tools buttons in the chat input are replaced by a single Tools button with a dropdown. Each item opens the same flow it did before, the button shows when a tool is active, and the dropdown works with the keyboard. The sources dropdown now opens one merged, searchable list of every campaign and dataset, with tabs and toggles to narrow it. After you select a campaign, the chat prompts you for your question.

  • A refreshed message layout

    • The response bubble now matches the message design prototype. The Vurvey logo and the Freckle name are gone from the bubble, the hover card shows the agent or persona type, and an ellipsis menu replaces the lightning bolt with delete inside it and a new retry action. "Show Reasoning" shows a brain image, the step count and a chevron, and each step is a bullet point. When a conversation has no messages yet, a suggestion pill shows beneath the composer, and clicking it sends that text as your message.

  • Campaign analysis that counts instead of guessing

    • Chat answers about campaigns are now exact and complete. Here is what changed, item by item.

    • Reading the whole campaign

      • Before: customers attached a campaign and chat often answered from the first page of responses as if it had read everything.

      • Now: chat reads the full campaign and tells them what it covered, for example "Based on 100 of 100 responses." Survey tools report how much of a campaign is left to read. When a campaign or question is attached, chat answers from the stored AI summaries first and pages through raw responses only when the summaries cannot answer.

    • Getting counts right

      • Before: "how many people chose X" got an estimate from a sample.

      • Now: choice counts and distributions come back exact and match the results page. Counts that span several questions, literal text matches and percentages come from real calculations, not from the model reading rows and counting in its head. The "quote respondents exactly" rule now reaches the model in every mode, and the prompt separates what respondents said they plan to do from what they did.

    • Asking about one question

      • Before: selecting a single question still pulled in answers and citations from the rest of the survey.

      • Now: chat stays on the question they picked. Survey tools respect the selected questions and give per-question counts and summaries.

    • Big open-text campaigns

      • Before: on large campaigns chat ran out of room partway through and answered from whatever it had read.

      • Now: it works through large campaigns in batches, so the answer reflects the whole campaign. Each batch gets bounded notes: themes with counts, verbatim quotes with response IDs, and outliers. The final answer is written from all notes, so every response is covered instead of a subset.

    • Citations they can trust

      • Before: answers came with a pile of "Reference 1, 2, 3..." links, many of which had nothing to do with the claim.

      • Now: citations point only at the evidence behind the sentence. Each turn keeps an inventory of everything shown to the model, and only entries that support at least one claim are saved as grounding and shown as citations. Answers built from counts or summaries cite the question-level source, not every response. The "Reference n" catch-all for workspace evidence is gone.

    • One caveat: this covers campaigns attached in chat.

  • Better dataset answers

    • Dataset searches no longer fail when the model sends numbers as text.

  • Smoother logout

    • Logging out no longer flashes a white page or does a full refresh. If the logout request fails from the sidebar dialog, you now see "Failed to logout, please try again" instead of a silent redirect.

  • Simulation progress shows more concepts

    • The concept display in the simulation progress view is taller, so a simulation with more than two concepts shows them in the container.

Bug Fixes

  • Chat

    • The "Conversations" breadcrumb and the export menu now show as soon as you start a new conversation, and they stay after a refresh.

    • There is no longer a loading avatar when no agent or persona is selected.

    • The Stop button now shows on every turn while a send is pending, not only the first turn, and a conversation no longer sits on "Loading..." forever.

    • Stopping a turn while a tool call is running now shows the "Stopped before the answer finished" note, and the tool shows as stopped instead of Running, including after a reload.

    • Leaving a chat while a response is streaming no longer leaves the response empty when you come back.

    • New conversations now appear in the recent conversations list as soon as they are created, with a placeholder title.

    • Chat no longer shows "Image unavailable" when the model mistypes a generated image URL.

    • The image editor no longer throws an error when you use the "remove" function.

  • Campaigns

    • "Download Chart" now works for the Topics spider chart on closed campaigns.

    • Longer answers in ranking question results are now fully visible.

Did this answer your question?