Feedback from Palma & Fontana on V3

Feedback following a call this morning with Me Fontana (Palma & Fontana law firm). Since the release of V3, they’ve been quite unhappy because they feel there have been several regressions, especially around the dictation/export workflow. Their main issue is the removal of the direct email export button after a dictation. Previously, they could send a dictation to their mailbox in one click so that secretaries could immediately process and format it. Today, they feel the workflow has become much more manual and cumbersome. As a result, they and some associates have gone back to using their old dictaphone/workflow for dictations. Main needs identified: bring back a simple and fast email export flow ideally add a “Send to secretary” type button maintain a smooth workflow for secretaries who do not use Whisperit Positive point: they really like the new legal assistant/research features in V3. The frustration is mainly focused on the dictation → email export workflow.

Calendar Sync

Inability to synchronize Outlook calendar with Whisperit calendar. This functionality is currently not available but is planned for release in the coming weeks.

Starter Templates not stored anywhere.

When I write a document there is a possibility to save it as a starter template. Then I look in templates and I can't find that template.

File upload to case fails due to infinite prompt

The customer reports an issue when attempting to upload a file to case `215a7877-66bb-44fe-a771-3a2e7f097d10`. They state that the system repeatedly asks the same question, preventing the file from being successfully uploaded. https://whisperit.ai/protected/cases/215a7877-66bb-44fe-a771-3a2e7f097d10 Thought: This user uses a lot of OCR - is the upload related to his limits?

{BUG} AI Bulk Rename: Downloaded Files Do Not Retain New Names

Customer reports an issue where files renamed in bulk by the AI do not retain their new names upon download. While the AI successfully renames the files within the application interface, the downloaded files revert to their original filenames. The customer has provided a link (https://www.swisstransfer.com/d/e3f5a9d7-5af9-44b9-b92a-e4f0d21c0a49) to illustrate the problem. The admin was initially unable to reproduce the issue and requested the specific command or message used for the AI renaming. User sends video.: https://drive.google.com/file/d/1oxry4YaC_DyegU2FtTEr3A-4kaRI6LmV/view?usp=sharing As when reproducing this error - the assistant gets stuck after the confirmation steps and I can't continue.

Whisperit Assistant: Incomplete Responses for Specific Input

The customer reports that the Whisperit assistant is unable to generate complete responses for a specific query, instead stating it cannot assist. The customer attempted basic troubleshooting steps including logging out and refreshing the page without success. Further testing by the customer indicates the assistant functions correctly with other inputs, suggesting the issue is specific to the provided information. The customer speculates this behavior might be related to the sensitive nature of the case being discussed (fictional resolution of a real, known, sensitive case). The URL for the affected chat has been provided: https://whisperit.ai/protected/assistant/5166b994-f63b-4db8-96d6-344c613abc7f.

Legal Search: Stop button non-functional

Customer reports that the 'stop search' button is not working when performing a 'Legal research' or 'deep search'.

The recent assistants on a case still not working properly.

The part where they are visible they are very small - they should be part of the summary of the case

We cannot send emails as @whisperit.ai

When I search in my email I only see the recent posts

I was trying to find my ticket for the event on my email and the search function didn't find it. (2 weeks old) I had to go to google, and open it from there.

Case: Detect existing files at upload & avoid AI reprocessing

Context Whisperit currently has no mechanism to detect when a document already exists in the system, either at upload time or at AI processing time. This is a root cause of our quota crisis: the user delete entire folders (30-60 documents) and reprocess everything from scratch because they have no visibility on what is already indexed. Direct consequence: runaway upload + Azure processing costs, and the drastic quota measure taken two weeks ago. Scenario: a lawyer builds an AI context on ~100 case files, generates a summary, returns 20 days later to redo the task. In the meantime, 10-30 new files have arrived in their SharePoint. Today, the only safe option is to delete the whole folder and re-upload everything. Scope Two layers, shipped together: Upload layer — duplicate detection On upload, detect duplicates based on name + size + content hash. Block silent duplication; surface explicit user choice: skip / replace / keep both (auto-rename). Apply at single-file and bulk-upload entry points. AI layer — incremental context awareness When the user adds files to an existing folder/context, the AI must recognize which documents are already in its context. Prompt the user: "I already have files X, Y, Z in this context. Integrate only the new ones (A, B, C) or rebuild from scratch?" Process only deltas by default; full reprocess on explicit request. Acceptance criteria Uploading a file already present in the folder triggers a duplicate dialog (skip / replace / keep both). Hash-based detection works even if the file is renamed. Adding files to a folder with an existing AI context only processes the new files unless the user opts for a full rebuild. Telemetry: track duplicate-detection hits and reprocess-avoided events to measure cost impact. Regression: existing single-file workflows are unaffected.

Case Assistant repeatedly crashing

The customer reported that the assisant in the case is repeatedly crashing, stating (It keeps crashing). They confirmed that the same error is occurring. The screenshot indicates the following error “Something went wrong while processing your request. Please try again. If this continues, contact support@whisperit.ai for assistance” The case id is https://whisperit.ai/protected/cases/60bcc16a-3b48-4b36-b9cc-e136f0088371

French UI localization

Make the Whisperit interface available in French for end users.

Export Word: When exporting, some texts appears 2x

As a user, when I export my Whisperit document to a Word document, the text in numbered list items is duplicated. Additionally, the numbering format is broken.

Unresponsive Chat Thread & Performance Slowdown Inquiry

Customer Jérôme reports that a specific chat thread (URL: https://whisperit.ai/protected/assistant/0261e881-3974-4e96-a5a7-cdcc0020798e) has been unresponsive for several days. The customer initiated a new chat due to the length and unresponsiveness of the original thread. Jérôme also notes that the system experiences significant slowdowns with dense chats and requests information on upcoming technical improvements in this regard. The customer asks if Whisperit can refresh the page on their side.

Google Drive Integration

To connect Whisperit's legal research tool directly to a customer's Google Drive. This will allow users to conduct deep, contextualized research across their own secure data by selecting specific folders or files within their Drive to serve as a private knowledge base for the AI's legal research and analysis features. (Customer based request)

Split a large PDF into small pdfs

As a user, I want to split a PDF into multiple documents so that I can better work with those files. It will allow me to better contextualize each file and also work with them using AI. Currently, I either use Adobe or iLovePDF that allows to either split or group pdfs (yet I’m a bit concerned about the security even thought it’s in Italy). I’d like to split it between specific pages. That’d be extremely useful.

Audio transcription fails for long recordings

User Story: As a user, I want long audio recordings to transcribe successfully, so that lengthy recordings can be processed without errors. Acceptance Criteria: Audio recordings longer than 30 minutes can be submitted for transcription. Large supported audio files do not fail with a server error during upload or processing. Transcription completes successfully for audio files that previously failed because of their size. Shorter audio recordings continue to transcribe successfully. Test: Upload a supported short audio recording and confirm transcription completes successfully. Upload a supported long audio recording over 30 minutes and start transcription. Confirm the upload is accepted and processing starts without an error. Confirm transcription completes and returns the expected transcript. Repeat with another large recording to verify the issue is resolved consistently. Technical note: The failure appears to affect the live dictation path only, not the import or meeting transcription paths. The live dictation flow sends audio inline to /api/voice/transcribe and does not switch to blob upload for larger files. Recordings over the request body limit can fail while parsing multipart form data, which leads to server errors for larger audio inputs.

Featurebase: sanitize page titles in v3 to avoid leaking names/IDs in dashboard recent views

Problem In the Featurebase dashboard, under recent page visits, page titles show raw route segments (including dynamic names/IDs) because v3 dashboard pages don't set document.title. Featurebase's SDK falls back to the URL path, which is how names/titles end up surfaced in the Featurebase admin view. Pointed out by Laura in #dev (Slack thread 2026-04-24): "Featurebase reads the page title we give to our pages, but we are actually not giving titles to pages on v3." Goal In Featurebase dashboard → recent page visits, each entry should display something like Whisperit · {sanitized title} — never a raw user name, doc name, client name, or other PII. Fix in a single file (messenger component) — do NOT add metadata.title to all 30 v3 route files. Ankan's constraint. Metadata on individual pages already looks fine; only the Featurebase-visible title needs to change. Proposed approach In dashboard/app/.../featurebase-messenger.tsx: Use usePathname() to get the current route. Sanitize it: strip UUIDs, numeric IDs, and any dynamic segments that could contain PII. Set document.title = \Whisperit · ${sanitized}`on every route change (viauseEffectonpathname`). Re-trigger Featurebase's changeLocale/identify (or equivalent) so the SDK picks up the new title on SPA navigations. Example sanitizer: const sanitized = pathname .replace(/\/[0-9a-f-]{36}/gi, '') // strip UUIDs .replace(/\/\d+/g, '') // strip numeric IDs .split('/').filter(Boolean) .map(seg => seg.charAt(0).toUpperCase() + seg.slice(1)) .join(' · ') || 'Dashboard'; document.title = `Whisperit · ${sanitized}`; Security / privacy requirements No PII in titles: no client names, document names, user names, emails, IDs. Whitelist of allowed route segments if possible; everything else becomes a generic label. Verify in Featurebase dashboard (recent page visits) after deploy that titles are clean. Acceptance criteria [X] Single-file change (messenger component only) [ ] Featurebase dashboard recent page visits show Whisperit · {sanitized-route} format [ ] No UUIDs, numeric IDs, or user-provided strings visible in any Featurebase-surfaced title [ ] Titles update correctly on client-side navigation (not just hard reloads) [ ] QA: manually inspect 5+ routes (incl. dynamic ones like /documents/[id], /clients/[id]) in Featurebase admin Context Slack thread: https://whisperit.slack.com/archives/C07BDM0LY14/p1777034773852469 Related: earlier Featurebase HMAC work (/api/featurebase/hash route) — keep existing auth flow intact.

Assistant: Mention "@" takes 20 seconds to retrieve data

As a user, I cannot use the “mention” as it takes more than 20 seconds to retrieve the data

Lost Audio - Recording & Diarization Missing; Sort by Date Not Working

User ID _ fe0cbd7d-9f50-4b1a-abc7-d71f0d250efe Customer reports that a recently recorded conference, approximately two hours in duration, is not appearing in 'My Vault'. The recording's diarization is also missing. Customer has confirmed that nothing is visible in 'My Vault' after checking. Additionally, the customer states that the 'sort by creation date' functionality is not working.

assistant issues

**Assistant Ignoring Case Documents:** When using the assistant within a case to summarize or draft responses, it requests information (e.g., case description, dispute object, legal points, exchanged documents) that is already present in the attached case documents. This indicates the assistant is not effectively utilizing the provided case context.