What we shipped recently

Fixes

• Exact numbers and codes are found — Knowledge search finds passages that contain a specific order, ticket or part number, even when the question is a full sentence. Passages also rank higher when the question and the document are in different languages.
• Fewer failed files — Files that hit a temporary outage while being indexed are retried for several hours instead of being marked as failed.
• Outpost folders show their real state — A shared folder that was moved, deleted or whose drive was disconnected now shows as unavailable, and "last sync" means the last complete scan. Deleting a folder also removes its content from keyword search.
• Replaced files are read again — A file that is replaced on an Outpost while it waits to be indexed is read again, so search results always match the current file.
• Clear errors from the embeddings API — When the API rejects an input, for example because it is too long, it now answers with a 4xx error that names the problem instead of a server error.

Faster search

Knowledge search runs on a new index and is up to 15× faster, so answers from knowledge sources arrive noticeably sooner.

SharePoint sync works again

SharePoint sources connected through a Teams site address or a link into a document library sync again. Since the previous update they had stopped picking up changes; no content was lost.

Scheduled tasks stay with their person

Scheduled tasks run, and show their results, only for the person they run as. Colleagues who edit the assistant still see a task's status and can switch it off or delete it.

Knowledge search in every workspace

Chat searches all available sources at once and shows each match with its file, page and section. Cmd+K finds files by name and can be filtered by source, file type and date.

Pictures in PDFs become searchable

Search now also finds pictures and drawings in PDFs, and answers show them directly.

Owners for data sources

Administrators can name people or teams who manage a data source without admin rights.

A more stable Outpost

Supports Windows Server 2016, works behind a proxy and starts with 200 GiB of storage.

More improvements

• SharePoint, OneDrive and Google Drive sync more reliably.
• Every Cloud source uses meaning-based search.
• The AI Cockpit shows how data sources are used.
• Generated images are marked as AI-generated.
• A MeinGPT account stays tied to the Google or Microsoft account it was first connected to.
• Numerous bug fixes.

New models: Claude Sonnet 5.5 and GPT-6.1 Sol

Claude Sonnet 5.5 and GPT-6.1 Sol are now available. Both are hosted in the EU and handle a context window of about 1M tokens, so long documents and extended conversations fit in a single chat. Administrators can switch them on in the model settings.

Knowledge search in chat

Chat can now search all connected data sources without selecting one first. Ask a question and the answer draws on every source available to you, ranks results across Cloud and Outpost, and shows which documents it found with links to the originals. Identical passages from multiple files appear once. Rolling out per workspace.

Documents in the search palette

The search palette (Cmd+K) finds files from connected data sources by name and opens the original. When nothing matches, a single Enter sends the question to a new chat that searches your data sources. Results and sources show whether a document is the current or an older version and when its source last synced; clicking a source opens the cited passage. Rolling out per workspace.

For administrators

• Honest progress and alerts — Data sources show the current processing step, searchable files, a rough time estimate, the queue position, and when a sync is stuck. Unreadable files explain why, and administrators get an email and in-app notification as soon as a source needs attention.
• Access review and permissions — Administrators see which data sources are shared with no one, and can share each with a team or everyone, or mark its sharing as reviewed. Source settings state plainly who can find which files; deleted files stop appearing in answers immediately, and access is recorded for security review.
• Improved connect flow — Adding a data source tests the connection first, names what went wrong and shows how many files it found. Sources are named at creation, duplicate folders are flagged, and reconnecting keeps existing settings and starts syncing right away.

More improvements

• Better German search — Knowledge search now finds German word forms, spellings with or without umlauts, and spreadsheet row contents such as part numbers. New data sources benefit right away; existing sources follow gradually as they are re-indexed.
• Gmail delete moves to Trash — The Gmail tool's delete action now moves emails to Trash, where they can be restored. The action is off by default and asks for the matching Gmail permission only when switched on.
• Scheduled tasks pick a valid model — New scheduled tasks start on a model that is actually enabled, and tasks whose model was switched off can be saved again.

Fixed

• Knowledge search prefers the current version of a document over archived or older copies. Answers name data sources by their display name, and files that cannot be previewed or downloaded explain why. On mobile, the sidebar no longer covers the chat.
• Outpost starts even when its search index exceeds the memory budget; keyword search remains available, and the console explains when semantic search is unavailable.
• Files that repeatedly fail during processing show a clear failure instead of staying pending indefinitely. Temporary file-access problems continue to retry, and updates to document processing give failed files another attempt.
• Semantic search in large Outpost folders uses smaller index sections that fit within the memory budget. Search results from smaller folders remain available alongside large collections.
• Outpost continues processing documents during sustained search traffic. Brief CPU spikes no longer pause processing; sustained high load still pauses it automatically.

Meeting summaries without taking notes

The MeinGPT notetaker joins Google Meet, Microsoft Teams and Zoom meetings, records them, and turns the recording into a transcript and a structured summary: topics, decisions, next steps and open questions. To send it into a meeting, paste the meeting link under Meetings → New meeting → Invite bot, add bot@meingpt.com to the calendar invite, or connect a Google or Microsoft 365 calendar so it joins scheduled meetings automatically.

Summary by email to everyone invited. After a recorded meeting, the organizer and the invitees can automatically receive the summary by email, so nobody has to write and send minutes. Administrators choose who receives it, and everyone can narrow this further for their own meetings.

The notetaker is available to every workspace that has signed version 2.0 of the MeinGPT data processing agreement. Administrators sign it under Settings → Contracts & DPA. Workspaces already using the notetaker keep it as is.

Models per team and person

A model that only one department needed used to be switched on for the whole company, or not at all. Now each model can be enabled or disabled per team and person: for example, a specialized coding model only for the development team, or everyone except the trainees. Administrators set this on the AI Models page or directly on a team's page, and each member's page shows which models that person can use. Image models work the same way.

Additionally, Apps, features and connectors can also be limited to the teams and people who need them, including who may let a connector write, for example send emails or create tickets. On top of that, workspaces can decide who may create and share assistants, projects, workflows, automations and API keys, or publish them for the whole company. This is switched on per workspace by the MeinGPT support team. When a member is missing something, their page explains why. (This feature is currently in beta and can be enabled by the support upon request)

More improvements

• Data source status at a glance — The data sources page shows which sources need attention and what to do next. Each source's page explains why a sync failed or a file isn't searchable, shows indexing progress, and offers a test search to see what a colleague would find.
• Settings are easier to find — Settings are grouped by topic. Models, apps and connectors sit side by side, the personal data filter has its own page, and API logs are part of the audit log.
• Simpler data source settings — Options that did not affect search results are removed.
• Outpost search in protected folders — On Windows, Outpost live search returns the files a person may read, even when the folder also contains files they cannot open.

Fixed

• Deleting a knowledge source no longer gets stuck, and crawl filters for websites are applied as configured.
• Answers list only the sources they actually cite.
• Syncing a knowledge source stays possible while it is still being indexed.

Fixes

• PDF citations open the cited page — Citations with page information open the PDF directly on the cited page. Original files download through the browser with their original filenames. Documents indexed before this release show the page once they are processed again.

For administrators

• Honest web crawler results — Web crawler sources whose start address cannot be reached now show a failed sync with the network cause instead of reporting success.
• Visible skipped files — Sync logs now list files that were skipped because their type is not processed for the source, such as images.
• No more empty upload sources — Adding a data source no longer creates an empty, unusable "manual upload" source.
• Clear messages instead of blank pages — Opening a deleted data source, or a data source page without admin rights, now shows a clear message with a way back instead of a blank or endlessly loading page.

Fixes

• Reliable regional S3 syncs — S3 data sources using regional AWS endpoints synchronize without requiring a separate region setting.
• Readable Word documents — Word files that appeared as archive metadata can now be indexed with their actual text.
• Automatic recovery for scanned PDFs — Temporary image-reading outages no longer leave scanned PDFs permanently unreadable. Processing resumes automatically after recovery.
• More resilient Microsoft connections — SharePoint and OneDrive connections retain their existing permissions during token refreshes and better tolerate slow responses.
• Automatic retries during high load — PDFs delayed by temporary processing capacity are retried instead of being marked permanently unreadable.
• Reliable large SharePoint libraries — Large libraries continue syncing without hitting the previous manifest capacity limit.
• Reliable Markdown indexing — Markdown files containing literal ampersands are indexed without rejecting valid source text.
• Stable processing of malformed documents — Pathological parser output is isolated before it can interrupt other document-processing jobs.

Assistants improve their own skills

A skill that no longer quite fits used to mean opening its files and editing them by hand. Now a sentence in the chat is enough: describe what should change, and the assistant rewrites the skill itself — it edits files in place, adds new ones or removes what is no longer needed. The skill keeps its name and stays available as before.

A new data processing agreement

Version 2.0 of the MeinGPT data processing agreement is here, with updated audit safeguards and a current, versioned list of subprocessors. Administrators can review, download and sign it directly in the privacy settings, or archive an agreement already signed on paper, and follow its status and download the signed evidence at any time. Existing contracts stay as they are until version 2.0 is accepted.

Signing moves the workspace to EU-hosted models. Models the new subprocessor list does not cover are withdrawn when the agreement is signed. Before signing, administrators see which models are affected and choose what replaces them; assistants and workflows are switched over automatically in the background.

Processing in the EU also tightens elsewhere: requests to Mistral models, including Mistral's document reading, now run in Mistral's EU region only, and the voice agent processes conversations in OpenAI's EU region only. Mistral Medium 3.5 and CodeStral cost 10 percent more per token as a result, in line with Mistral's EU pricing.

More accurate knowledge search

Knowledge search finds the right passage more often. It returns more accurate matches, keeps the relevant surrounding text with each result, and finds matching files even when they are spread across several result pages.

When something goes wrong, it now says so: a failed search is reported as a failure instead of looking like a search without matches, and missing files or unreadable text are named clearly instead of appearing as empty documents.

More improvements

• Planner task comments — The Microsoft Planner connector can read a task's comments and post new ones, so a task discussion continues from the assistant without switching to Planner. No additional Microsoft permissions are needed.
• Pasted links keep their address — Links pasted from Outlook, Teams, SharePoint or a web page keep their web address in the chat input and in workflow steps, so the assistant sees the actual link instead of only its title.
• Cheaper meeting recordings — A notetaker recording now costs €0.80 per audio hour instead of €1.00, including the transcript and the standard summary.
• Notetaker in signed-in Teams meetings — The notetaker joins Teams meetings with a signed-in account when one is available, so it can also record meetings that don't admit guests.
• Workshop slides by email — Workshop organizers can attach slides, and attendees receive them by email an hour after the workshop ends.

Fixed

• Scanned PDFs import correctly, and details for empty files open without errors.
• Data sources outside the default cloud storage update their file counts after a synchronization, and synchronization times reflect completed runs.
• The Google Drive folder picker shows every folder, even on large drives.
• The notetaker's "Join now" works for meetings that are already running, moving a calendar event no longer hides its earlier recordings, and recordings without speech are explained clearly instead of failing repeatedly.

For administrators

• Perplexity Sonar retired — Saved assistants and workflows that used Perplexity Sonar move to MeinGPT Web automatically. External integrations that call Sonar through the API must switch to an assistant with web search before the announced cutoff.
• Google Drive folders — The folder picker lists folders in a consistent order, can be searched by name, and can connect a whole shared drive or an entire My Drive.
• Usage overview — Sidebar and profile share one compact usage overview. In workspaces without personal limits it stays hidden until an administrator switches it on, and it is always hidden in All-inclusive workspaces.
• Outpost services in Tool Management — Registered Outpost databases and MCP services show when they still need to be enabled in Tool Management, and members see why a service is missing from their assistants.
• Outpost on Windows Server 2019 — The Outpost now reads documents on Windows Server 2019. The installer no longer accepts Windows Server 2016, and a failed installation names the missing Windows component instead of failing file by file.
• More robust Outpost — A failing service no longer affects the registration of the others; the affected service shows its failure and a recovery step in the desktop app. Semantic search stays accurate and available while Outpost indexes a large folder for the first time.

Five new models

Five models join the selection this release, all hosted in the EU.

• GPT-6 Sol and GPT-6 Luna run on Azure in West Europe, both with a one-million-token context window — long enough for large document sets in a single conversation.
• Claude Opus 5.5 runs on Google Vertex in the EU and joins the rest of the Opus line as the strongest option for demanding reasoning work.
• Gemini 3.7 Flash and Gemini 3.8 Flash run on Google Vertex in the EU, also with a one-million-token context window, and are the fast, inexpensive end of the selection.

All five are switched off by default. Administrators enable them per workspace under Settings, Models.

Gemini 3.6 Flash also gets cheaper. It now bills at Google's promotional rate, which cuts its credit cost by about 45 percent. Existing conversations using it consume fewer credits from the next run onwards.

Meeting recordings

The meeting notetaker gets the two things that were missing around it: an end, and a recovery path.

Recordings can now end early. Stop and save makes the notetaker leave the meeting while the content captured up to that point is still processed into a transcript and a summary — rather than the choice being between letting it run to the end or losing the recording.

Failed recordings stay visible. A recording that fails now remains in the list with an explanation instead of disappearing, and can be recovered whenever the audio or a transcript is still available. Status filters keep the list manageable, and a partial transcript that was saved remains readable.

Replies inside Gmail threads

An assistant asked to answer an email used to produce a separate new message, which then had to be moved or rewritten by hand. A reply can now be drafted and revised inside the existing conversation: the draft stays in the original thread and keeps its subject, so it reads as the answer it is.

Outpost and data sources

This release continues the indexing and connection reliability work.

Connected databases and on-premise connectors are recognised correctly when they are working, instead of being reported as empty or unavailable. Database and MCP tools stay reachable after registration, a restart and a recreated service, and a service that is offline names itself and a recovery step instead of showing a generic loading failure. During the first status read after a start, Outpost says that it is checking the connection, so a normal startup no longer looks like a connection problem.

Indexing recovers on its own more often. Interrupted indexing resumes after an update or a crash instead of waiting up to 30 minutes before search can finish rebuilding. When an Outpost installation cannot extract files at all, it stops retrying every document, reports a clear installation error, and picks indexing back up once the installation is repaired. On macOS, a rolled-back update leaves the previous version launchable.

In the cloud vault, existing search results stay available when an update fails, and interrupted processing resumes without doing the same work twice. Document extraction, website synchronisation and source downloads recover more reliably from failures and cancellation.

More improvements

• Selected data sources are searched first — A data source picked in the chat input is now searched before the answer is written, instead of the chat asking where the documents are stored. File name searches also find a file when the words are in a different order than in the file name.
• Assistant folder restrictions stay local to the assistant — Folder and site restrictions saved inside an assistant apply only to that assistant, and leave personal connection settings unchanged.
• Clearer AI errors in Buddy — When an AI response fails for a recognised reason, Buddy explains what happened and offers a help link and a reference that can be copied into a support request.
• Workshop invitations — Calendar entry and emails now use one consistent name, and the calendar entry explains what the workshop covers and how to sign in beforehand.
• New icon — MeinGPT Outpost and the Outlook add-in now show the new MeinGPT icon.
• A failing database connection no longer takes Outpost with it — Folder indexing and search stay available when a configured database connection fails, and a connection check names the affected service and the cause of the failure.
• Faster Outpost rollouts — Updates start sooner once a release reaches a device's ring, and completed installations appear in the campaign status within minutes.

Events in Learn

Workshops used to live in a separate admin setting, which made them hard to find for the people actually attending them. Learn now has an Events area instead: upcoming workshops are listed with their dates, registration takes one click, and attended sessions stay visible as a personal overview.

Admins manage the whole catalog from the same place — sessions, trainers, meeting links and email defaults, plus participants and attendance per event. Attendance can be imported from a CSV, sessions can be cancelled or rescheduled, and participants can be notified when an event changes. This replaces the old workshop settings screen.

Meetings as chat attachments

Meeting transcripts previously had to be found through a general search. A specific meeting can now be attached directly in a chat through the + menu or an @ mention, and the picker is searchable. Meetings with the same name are distinguished by date and time, so the right one is attached the first time.

Model phases and dates

Providers retire their models, and until now that only became visible when a model stopped working. Model settings now show whether a model is in preview or being phased out, the exact dates involved, and which model replaces it.

Models that are being phased out can no longer be picked for new assistants and workflows, and retired models disappear from the pickers and from settings entirely. A new page in the admin guide explains the phases.

Outpost and data sources

This release is a large round of indexing reliability work.

Multi-page scanned PDFs are read and indexed instead of failing over and over, and a file that still exceeds a local processing limit now says so rather than asking for a valid file to be replaced — and it no longer holds up the rest of the folder. DOCX and PPTX files recover from a failed extraction worker instead of leaving indexing stuck on a generic error, and persistent failures name the underlying cause. When text recognition cannot start at all, PDFs report the actual reason and are retried once it works again.

Very large or slow folders now finish their initial scan: each pass continues where the last one stopped, files are indexed along the way, and the console reports a large folder being scanned in steps instead of a timeout. Scans that were cut short complete reliably, an unfinished scan is shown as an interrupted scan that retries rather than a broken file, and Outposts no longer report frequent processing failures when nothing failed. A network folder that stops responding no longer freezes indexing.

In the cloud vault, a failure on one file no longer disrupts the others. Complex PDF diagrams and small labels are read more reliably, WebDAV folders sync correctly, and a Microsoft access restriction is now told apart from a processing error. Existing search content stays available when an index update fails, interrupted processing can resume, and a stale processing attempt can no longer overwrite a newer version of a file. Deleted files disappear from file listings, while empty files and files still waiting to be indexed remain visible. Sync runs also count file sizes, extracted text and text sections once instead of twice.

More improvements

• Microsoft SQL Server — The database tool returns query results, table lists and column details for SQL Server connections, and reports how many rows an insert, update or delete changed.
• Large files in workflow API runs — Workflow runs started through the API can read large file attachments instead of coming back empty, and scanned PDFs and photos reach the model reliably. A step that still produces nothing returns a clear error instead of an empty response.
• Shared assistants — A shared assistant can use a custom MCP server without a separate sign-in when the server requires no authentication.

For administrators

• Contract credit packages — Contract credits now arrive automatically at the start of each contract period. Promotional allowances show their schedule and top up eligible workspace balances, without unused monthly grants piling up.
• Notetaker appearance — The greeting the meeting notetaker posts when it joins can now be set, and its avatar picked from a set of MeinGPT images, or a custom image where that is enabled.
• Outpost console on Windows — The console opens Windows service status without administrator approval; installation, repair, service controls and configuration changes ask for approval only when they need it. If it cannot open with administrator rights at all, it now explains how to continue without them and can reopen itself as the standard status view, instead of showing a crash dialog.

Fixes

• Videos, audio recordings, document previews and downloads could stop loading until the affected server restarted. The connections to the file storage are released again, and requests to it no longer wait indefinitely.
• Workshop invitations and registrations work again, and opening session details dims the workshop dialog behind them.
• The AI Connect page no longer advertises the expired early-bird discount.