AI-Studio/runtime/packaging/linux/org.mindworkai.AIStudio.metainfo.xml

<?xml version="1.0" encoding="UTF-8"?>
<component type="desktop-application">
  <id>org.mindworkai.AIStudio</id>
  <name>MindWork AI Studio</name>
  <project_license>FSL-1.1-MIT</project_license>
  <metadata_license>MIT</metadata_license>

  <summary>MindWork AI Studio is a free, independent cross-platform desktop app for local and cloud LLMs across providers, built to democratize AI access.</summary>
  <developer id="org.mindworkai">
    <name>MindWork AI Community</name>
  </developer>
  <content_rating type="oars-1.1" />
  <description>
    <p>
      MindWork AI Studio is a free desktop app for macOS, Windows, and Linux. It provides a unified user interface
      for interaction with Large Language Models (LLM). AI Studio also offers so-called assistants, where prompting
      is not necessary. You can think of AI Studio like an email program: you bring your own API key for the LLM of
      your choice and can then use these AI systems with AI Studio.
    </p>
    <p>Key advantages:</p>
    <ul>
      <li>
        Free of charge: The app is free to use, both for personal and commercial purposes.
      </li>
      <li>
        Democratization of AI: MindWork AI Studio runs even on low-cost hardware, including
        computers such as Raspberry Pi. This makes the app and its full feature set accessible
        to people and families with limited budgets. You can start with local LLMs or use
        affordable cloud models.
      </li>
      <li>
        Independence: You are not tied to any single provider. Choose the providers that best
        suit your needs, including OpenAI, Perplexity, Mistral, Anthropic, Google Gemini, xAI,
        DeepSeek, Alibaba Cloud, OpenRouter, Hetzner, IONOS, LiteLLM, Hugging Face, Groq,
        Fireworks, Helmholtz, GWDG, and self-hosted models.
      </li>
      <li>
        Assistants: Use ready-made assistants for common business and other tasks without writing prompts yourself.
      </li>
      <li>
        Unrestricted usage: Unlike services that impose limits after intensive use, MindWork
        AI Studio lets you use provider APIs without restrictions imposed by the app.
      </li>
      <li>
        Cost-effective: You only pay providers for what you use, which can be cheaper than a
        monthly subscription when used infrequently. For intensive usage, API costs may be
        higher, so you should monitor your provider accounts and use prepaid credit or cost
        limits when available.
      </li>
      <li>
        Privacy: Control which providers receive your data using provider confidence settings
        and assign different protection levels to different tasks.
      </li>
      <li>
        Flexibility: Choose the provider and model best suited to your current task.
      </li>
      <li>
        No bloatware: The app requires little storage and memory and has minimal impact on
        system resources and battery life.
      </li>
    </ul>
  </description>

  <launchable type="desktop-id">org.mindworkai.AIStudio.desktop</launchable>

  <categories>
    <category>Utility</category>
    <category>Office</category>
    <category>Science</category>
  </categories>

  <keywords>
    <keyword>AI</keyword>
    <keyword>Assistant</keyword>
    <keyword>Privacy</keyword>
  </keywords>

  <url type="homepage">https://mindworkai.org</url>
  <url type="bugtracker">https://github.com/MindWorkAI/AI-Studio/issues</url>
  <url type="contact">https://github.com/MindWorkAI</url>
  <url type="contribute">https://github.com/MindWorkAI/AI-Studio#contributing-ov-file</url>
  <url type="vcs-browser">https://github.com/MindWorkAI/AI-Studio</url>

  <provides>
    <binary>mind-work-ai-studio</binary>
  </provides>

  <branding>
    <color type="primary" scheme_preference="light">#b4bed5</color>
    <color type="primary" scheme_preference="dark">#707e99</color>
  </branding>

  <screenshots>
    <screenshot type="default">
      <image>https://github.com/MindWorkAI/AI-Studio/blob/main/documentation/AI%20Studio%20Home.png?raw=true</image>
      <caption>Getting started</caption>
    </screenshot>
    <screenshot>
      <image>https://raw.githubusercontent.com/MindWorkAI/AI-Studio/refs/heads/main/documentation/AI%20Studio%20Assistants.png</image>
      <caption>Assistants</caption>
    </screenshot>
  </screenshots>

  <releases>
    <release type="stable" version="26.9.1" date="2026-09-27">
      <description>
        <ul>
          <li>Added a way to copy an entire chat, either with the button in the chat toolbar or next to the chat in the chat list. The copy opens right away so you can continue in it, while the original conversation stays exactly as it was. Many thanks to Peer Hogeterp (<code>peerschuett</code>) and Jens Erler (<code>j-erler</code>) for this feature.</li>
          <li>Added a way to roll a chat back to an earlier AI response. The response you pick stays, and every message after it is removed permanently, together with the attachments of those messages.</li>
          <li>Added a way to save a single code block of an answer. When an answer holds a web page, a LaTeX document, or a Markdown text, the export menu now offers that block as a file of its own.</li>
          <li>Added tools that AI models can use on their own, starting with Web Search and Read Web Page. When you ask something a model cannot answer from what it knows, it now searches the web, reads the pages it found, and answers with the sources it used. You decide which tools a model may use, right below the message field, and you can watch it work: AI Studio shows which tool is running and, afterward, every call it made with its result. Whether tools are offered at all depends on the model because it has to support them. Read Web Page works right away; for Web Search you pick a search service in the app settings — Tavily or Staan with a free API key, or a SearXNG instance you run yourself. Set up more than one, and they can take turns when one of them finds nothing, or be asked all at once with their results combined. Many thanks to Peer Hogeterp (<code>peerschuett</code>) and Nils Kruthoff (<code>nilskruthoff</code>) for building this feature.</li>
          <li>Added answers that appear word by word even while the AI uses its tools. You read along as the model writes, including the short note it puts down before it looks something up, and the answer that follows a tool call arrives the same way instead of all at once at the end.</li>
          <li>Added safeguards around everything these tools bring back. Anything fetched from the web is treated as untrusted: AI Studio removes instructions hidden in a page before a model reads it and tells you when it did, exactly as it already does for the documents and web pages you load yourself. A model can never point a tool at your own network, and Read Web Page opens only web addresses from your chat or from its tools unless you let it choose them freely. Each tool states how much you have to trust a provider before it may be used with it, so your questions do not travel further than you allow. You can adjust that requirement per tool in the app settings.</li>
          <li>Added tools to the assistants. Each assistant has its own tool settings: which tools it starts with and whether you get to change them while you work. The chat, the coding assistant, and the Slide Builder always show the selection; for every other assistant you switch it on where you want it.</li>
          <li>Added tools to the Batch Processing assistant, so a batch run can look things up while it works through your documents. You choose them next to the instructions of the job, and every document is processed with the same set. The log file now records which tools were used for each document, and whether a call failed or was blocked, so you can tell how an answer came about.</li>
          <li>Added tools to the policies of the Document Analysis assistant. A policy states which tools an analysis may use, and the AI uses exactly those — nobody has to pick them per document. AI Studio warns you beforehand when the provider you selected is not trusted enough for a tool the policy names. IT departments can roll policies out together with their tools.</li>
          <li>Added tools to assistant plugins and direct-chat launchers. Plugin authors name them in the new <code>ToolIds</code> field, either as the tools an assistant runs with or as the tools a launcher preselects for the chat it opens; the example assistant plugin shows both. Which tools an assistant asks for is part of what you get to see before you enable it: its security card names them, and the security audit takes them into account.</li>
          <li>Added tools to the Assistant Builder. For a direct-chat launcher you pick them yourself, alongside the workspace, provider, and data sources. For an assistant, the AI chooses from the tools installed here and says so in the draft, so you see the decision before the assistant is written.</li>
          <li>Added tools and data sources to your chat templates. A template can now decide which tools a chat starts with and which of your documents it may search, so a template such as "Research in our intranet" is complete on its own instead of leaving you to set the same things up by hand every time you switch to it.</li>
          <li>Added a way for IT departments to roll out chat templates that bring their own tools and data sources. You do not have to write any of it by hand: set the template up in the app, then export it as ready-made Lua code for your configuration plugin.</li>
          <li>Added organization-wide management for tools. Among other options, IT departments can switch tools off entirely, disable individual ones, or define the provider trust a tool requires. You do not have to write any of it by hand: set a tool up in the app, then export its configuration as ready-made Lua code for your plugin, with encrypted API keys if you want them.</li>
          <li>Added tool calling to the abilities you can state yourself in the expert provider settings. When you use a model AI Studio does not recognize as tool-capable, you can now declare that it is, the same way you already could for image input or reasoning. It works in the other direction as well: a model AI Studio has never heard of is offered tools, because most models can use them by now. Should one turn out not to be able to, AI Studio says so in plain words and points you to the same setting to switch the ability off again, instead of only passing the provider's error on.</li>
          <li>Added a Search Confluence tool, so the AI can look things up in your organization's Confluence Data Center wiki. Set your wiki's address in the tool settings. Selecting the tool also selects Read Web Page, which the AI uses to open the pages it found. Confluence Cloud is not supported yet. Many thanks to Peer Hogeterp (<code>peerschuett</code>) for this tool.</li>
          <li>Added sign-in with your operating system account to the Search Confluence tool for a wiki on your organization's network, so you do not have to enter a password. When your wiki does not accept that sign-in, AI Studio tells you so instead of reporting an empty search.</li>
          <li>Added safeguards to the Search Confluence tool: it works only with a High-confidence provider, and it never follows a redirect that leads away from your wiki. Each search appears in the sources of the answer.</li>
          <li>Added support for OpenAI's GPT-6 Astra, Sol, and Luna.</li>
          <li>Added a way to keep improving a prompt in the Prompt Optimizer. Select "Improve further" to move the latest proposal back into the prompt field, edit it the way you want, and optimize it again. Your recommendations and everything you selected stay as they are.</li>
          <li>Added the context window to what AI Studio knows about a model, wherever its metadata states one.</li>
          <li>Added a live read of that context window at the providers which report it, among them Mistral, Groq, OpenRouter, and self-hosted vLLM servers. You then get the window your own server was started with, not the one the model card advertises.</li>
          <li>Added a token count below the message field, so you always see how much of the conversation you have used. It counts everything that travels along: your messages, the files you attached, what your data sources contributed, and the tools you offered the AI. It can be an estimate when a provider does not give AI Studio everything it needs to count exactly.</li>
          <li>Added a warning when your conversation holds more images than the model accepts, wherever we know that limit. The Visual Briefing assistant stops before anything is uploaded, instead of letting the provider refuse it afterward.</li>
          <li>Added the context window and the image limits to the expert provider settings, next to the abilities you could already state there. Leave a field empty, and AI Studio keeps its own answer, which you see as the placeholder. IT departments can state the same numbers for the providers they roll out.</li>
          <li>Added model plugins, so IT departments can describe the models their organization runs itself.</li>
          <li>Added local RAG as a beta feature, so the AI can answer from your own documents. You point AI Studio at a folder or at a single file, and it prepares those documents in the background so they can be searched. Ask a question with such a data source selected, and only the passages that fit your question reach the model, along with where each one came from. We will keep developing it together with the people who use it: to try it, open the app settings, allow preview features down to beta, and then enable the RAG feature. Many thanks to Paul Koudelka (<code>PaulKoudelka</code>) for around ten months of work on the concept and the implementation.</li>
          <li>Added the setup for local data sources. You pick an embedding provider, and AI Studio asks for your confirmation before any document goes to a cloud service. It keeps up with your files as they change, shows the progress on a page of its own, and checks every document for hidden instructions before indexing it. Documents without readable text, such as scanned pages, are remembered as such, so AI Studio does not work through them again after every start — it comes back to them once they change.</li>
          <li>Added a way to open the sources of your own documents: click a source below an answer, and the document opens in the program your system uses for it.</li>
          <li>Added a jump to the right page for the sources of your own documents (RAG), so a PDF opens directly where the passage was found, wherever your system and its program support it.</li>
          <li>Added a way to show a source of your own documents (RAG) in your file manager.</li>
          <li>Added the details of the two databases behind local RAG to the information page: which versions they run, how much space they use on your disk, and how much they hold.</li>
          <li>Added a repair for your local data sources. Should the index of a data source ever become unreadable, AI Studio now says so instead of quietly finding nothing and leaves that source out of your chats until it works again.</li>
          <li>Added the repair itself as a button next to each of your data sources. Rebuilding an index sends your documents to your embedding provider once more, so AI Studio asks you first and never starts it on its own.</li>
          <li>Added Semantic Search: when your model can use tools, it now searches your data sources itself — only when a question calls for it, as often as it takes, and with search terms it works out from your conversation.</li>
          <li>Added a choice per chat of how your data sources are used: the AI searches them itself (Semantic Search), or AI Studio searches them with every message you send (classic RAG). Models that cannot use tools always use classic RAG, and the data source selection tells you when that happens.</li>
          <li>Added organization-wide management for how the data sources of new chats are searched. IT departments can make it a default or lock it.</li>
          <li>Added a check that recognizes your own tokenizer by its content rather than by its file name. Tokenizers are almost always named alike, so swapping one for another is now noticed, and you are asked about it.</li>
          <li>Added support for several drop areas on the same page. More complex assistants can now receive files or folders by drag and drop at more than one place.</li>
          <li>Added drag and drop to the input and output folder of the Batch Processing assistant: drop a folder onto either field to choose it.</li>
          <li>Added ways to load text from a file and drop zones for them, throughout the assistants and dialogs. We went through them one by one, so many fields that used to accept typed text only now take the content of a file as well.</li>
          <li>Added an optional API key to every server you host yourself, among them LM Studio, llama.cpp, and whisper.cpp. Such a server may ask for one itself or sit behind a login your organization placed in front of it. So far, only Ollama and vLLM could be given a key.</li>
          <li>Added a setting for the audio quality used when your speech and your audio and video files are transcribed. AI Studio prepares every recording before it goes to your transcription provider, and you now decide how much detail it keeps: a lower quality travels faster, a higher one gives the transcription model more to work with. You find it in the app settings, right below your transcription provider. Thanks, Dominic Neuburg (<code>donework</code>), for this contribution.</li>
          <li>Added organization-wide management for the audio quality used when transcribing. IT departments can set the quality their organization works with and lock it, or leave it as a default their colleagues are free to change.</li>
          <li>Improved loading web content in the assistants: it now uses the same reader as the Read Web Page tool, which extracts the main content of a page more reliably and skips navigation and boilerplate. Pages from your own network, including local servers, keep working as before. When a page cannot be read, AI Studio now says why instead of leaving the field empty.</li>
          <li>Improved the app icon. The previous one was generated by an image model; the new one was created based on it and keeps the familiar green landscape with the chat bubble. Because it is now a vector drawing, it stays sharp everywhere it appears: in your taskbar or dock, in the window list, and on the start screen while AI Studio is loading.</li>
          <li>Improved how AI Studio works out what a model can do. Every model family now stands on its own, together with the page it was read from, and our build refuses rules which contradict each other or name no source. That way, mistakes are caught before they ever reach you.</li>
          <li>Improved the list of your attached files: every file now appears under the folder it came from, and each folder is named only once, no matter in which order you attached your files.</li>
          <li>Improved organization-wide provider management: IT departments can now separately prevent users from adding chat, transcription, or embedding providers. The existing master setting still overrides all three provider-specific settings.</li>
          <li>Improved the provider selection throughout the assistants: when there is nothing to choose from, it now says why. Either you have not set up a provider yet, or none of yours is trusted enough for what you are doing. Before, the list was simply empty.</li>
          <li>Improved the settings of your embedding providers and data sources. Some of them decide how your documents are read, so changing one means preparing every document all over again. AI Studio now asks before that happens, names the data sources it would affect, and says when a cloud provider charges you for it.</li>
          <li>Improved the dialogs of your embedding providers and data sources: when a change would mean preparing all your documents again, and you decide against it, nothing is saved and the dialog stays open with your change in front of you, ready to be corrected.</li>
          <li>Improved what happens when you open an embedding provider whose server is unreachable or no longer offers the model you chose. That model stays selected, and AI Studio tells you the server does not have it right now. The documents you already prepared keep working, and you are not asked to prepare them again over a change you never made.</li>
          <li>Improved the question AI Studio asks before you delete an embedding provider. It now names the data sources depending on that provider, together with what they can still do without it.</li>
          <li>Improved what the AI is told when it answers from your own documents (RAG): it now learns which page a passage came from, so it can name the page an answer rests on.</li>
          <li>Improved what happens when you ask for web content to be cleaned up and no model is available for it. AI Studio loads the page and tells you it arrived uncleaned, instead of quietly handing you the raw page with its navigation and advertising still in it.</li>
          <li>Improved the file name AI Studio suggests when you export an answer. Instead of always proposing "export", it now suggests the name of your chat or of the assistant you are working in. In the Document Analysis assistant, it suggests the name of the policy.</li>
          <li>Improved the protection against prompt injection. It now also finds instructions disguised with the escape codes that formats like JSON and XML use, both in your own documents and in content from the web.</li>
          <li>Changed what a tile that opens a chat directly starts with: when it uses a chat template that brings its own tools or data sources, that template decides them. The Assistant Builder says so while you build such a tile.</li>
          <li>Changed which model a security check of an assistant plugin may fall back on. When you have set none aside for these checks, AI Studio uses the model you were working with, but only when that model meets the trust your organization requires for a check. One ruled out for that purpose no longer gets to read a plugin's code.</li>
          <li>Changed how provider trust and provider confidence work together. Marking a provider as trustworthy in a configuration no longer also satisfies a required confidence level: one says who runs the provider, the other how confidential it is. Organizations raise a provider's level in their own confidence scheme instead. This applies beyond local data sources, for example, when a model reads a page from your intranet.</li>
          <li>Fixed the abilities AI Studio assumed for many models. We checked the families against their documentation: some models gained image input, reasoning, or tool calling, others lost an ability they never had.</li>
          <li>Fixed the Document Analysis assistant freezing while you edited a policy. It needed a change of yours to be saved in the background just as you were making the next one — picking a provider, for instance — which is why it hit some of you again and again and others never at all.</li>
          <li>Fixed a renamed policy losing its new name in the Document Analysis assistant. The name was kept only when you happened to change something else afterward.</li>
          <li>Fixed model names that a provider writes in its own way not being recognized at all, such as the colon Ollama puts before the variant. Those models were treated as plain text models and lost every other ability.</li>
          <li>Fixed a model resold under a plain name not getting the abilities it really has.</li>
          <li>Fixed models showing up among the chat models, although nobody can chat with them, such as the ones that draw, film, compose music, transcribe speech, read scanned documents, or work through a task on their own. The same goes for models that need a live connection AI Studio cannot open.</li>
          <li>Fixed the model lists of your providers showing the wrong models. Some your provider offers were missing, among them Google's Gemma models, while others that no longer exist were still on offer, among them an older Grok version. The lists now follow what your provider reports.</li>
          <li>Fixed having to type the name of an embedding model by hand when you set up a provider on your own Ollama or vLLM server. AI Studio now asks your server which models it has and offers them in a list, the same way it has always done for LM Studio.</li>
          <li>Fixed the model list for transcription on your own server offering everything the server has, including models that can only chat or create embeddings. You are now offered the models that can actually transcribe, and nothing else.</li>
          <li>Fixed the model of a transcription provider on your own Ollama server not being saved. An empty model was stored instead, so transcribing with it could not work, and picking a different model changed nothing.</li>
          <li>Fixed a dropped file being processed several times, e.g., after the computer woke up from sleep.</li>
          <li>Fixed nothing happening when you dropped a file onto the list of your attached files. You can now add files to that list while it is open.</li>
          <li>Fixed the preview of an attached file ignoring dropped files. Drop another file onto the preview, and it is attached and shown right away.</li>
          <li>Fixed AI Studio shutting down without warning when two PDF files were read at the same time, e.g., when you previewed one while another was still being read in the background.</li>
          <li>Fixed the web address staying in the field when you reset an assistant that loads content from a web page.</li>
          <li>Fixed the web address being gone when you leave such an assistant and come back to it later.</li>
          <li>Fixed an assistant refusing to work after you switched on the cleanup of web content without having chosen a model first. Its button did nothing at all, and only closing the assistant and starting over helped. The cleanup now follows the model of the assistant you are working in, the moment you pick one.</li>
          <li>Fixed the Visual Briefing assistant (in preview) not scrolling, which put everything below the window edge out of reach and made the assistant unusable. The briefing preview is now shown at its intended size inside its frame, and switching between the desktop, tablet, and mobile view changes its width as it should.</li>
          <li>Fixed exported answers losing their sources. When an answer is based on web pages a tool read or on documents of your own, the exported file now lists those sources in every format AI Studio writes.</li>
          <li>Fixed the copy button leaving the sources behind. Copy an answer, and its sources come along.</li>
          <li>Fixed exported web pages showing a cryptic string of letters and digits as their title in the browser tab. They now carry the name of their file.</li>
          <li>Fixed a tile that opens a chat directly always demanding a workspace. Leave the workspace empty in the Assistant Builder, and the tile opens a disappearing chat instead.</li>
          <li>Fixed the same restriction for plugin authors: a direct-chat launcher can now open a chat without naming a workspace. The example assistant plugin shows both ways.</li>
          <li>Fixed the security check of an assistant plugin being impossible when no model is set aside for such checks, and you have no app-wide default either. The dialog now lets you pick one, and that choice applies to this one check. Before, the button to start the check was greyed out with nothing saying why, so the plugin could not be enabled at all.</li>
          <li>Fixed a security check that failed, leaving you no way to try again. When a check ends without a result, because the model could not be reached or a key was wrong, you can simply start it once more instead of closing the dialog and opening it anew.</li>
          <li>Fixed a security check that failed counting as a check that took place. Such a check no longer unlocks an assistant plugin for use, and it no longer replaces the last result that did say something about that plugin.</li>
          <li>Fixed data sources you picked for your chats vanishing from the selection without a word when they cannot be used. AI Studio now lists them by name, so you can see why an answer was created without them.</li>
          <li>Fixed the data sources you picked for a chat being forgotten the moment you changed your selection while one of them could not be used. Such a source stays selected and is used again as soon as it is available.</li>
          <li>Fixed the silence when the step that picks the fitting passages out of your documents fails. You are told that the answer rests on everything that was found.</li>
          <li>Fixed passages from your data sources still reaching the AI after you switched the data sources of a chat off. Likewise, when a search in them found nothing, the AI kept receiving the passages an earlier message had brought up.</li>
          <li>Fixed the regenerate button taking an answer away without producing a new one. This happened in chats started from a template that holds no question of your own.</li>
          <li>Fixed the counter above an answer, which shows how many sources it rests on, doing nothing when you clicked it. It now takes you down to the sources.</li>
          <li>Fixed the list of models staying empty at a server you host yourself, which made the model you had picked look as if it had vanished. Your key was there all along, it just was not read when the settings opened.</li>
          <li>Fixed AI Studio asking such a server for its models with an empty key attached when you had stored none at all. Servers behind a login turn those requests down.</li>
          <li>Fixed a key that could not be saved going unmentioned for the servers you host yourself. You are now told what went wrong, instead of the settings simply staying open.</li>
          <li>Fixed transcripts quietly losing what was said softly, such as a greeting at the very beginning of a recording. AI Studio compressed recordings so far before sending them to your transcription provider that the model could no longer make out those passages. Recordings now keep enough details for the whole of what you said to arrive.</li>
          <li>Fixed AI Studio seeming to hang for minutes when a chat had grown too large for the model. Some providers turn such a chat down in a way AI Studio did not recognize, so it kept sending the very same chat again and again. You are now told right away that the chat, including its attachments, is too large for the selected model.</li>
          <li>Fixed AI Studio trying for minutes when a provider turns a request down for good. Such an answer does not change by asking a second time, so AI Studio now stops at the first one and tells you what the provider said about it.</li>
          <li>Fixed errors about a provider arriving as two messages at once, the second of which spoke of several attempts that were never made. You now get the single message which names the cause.</li>
          <li>Fixed the button in the chat toolbar that deletes the current chat and starts a new one doing so without asking. It now asks for your confirmation first, just like the chat list does, because a deleted chat cannot be brought back. The button shows a delete icon in red now, instead of one that looked like a reload.</li>
          <li>Fixed AI Studio following your system into light or dark mode even though you had chosen a fixed color theme in the app settings.</li>
          <li>Fixed AI Studio keeping its previous color theme after your computer woke up from sleep, when your system had switched between light and dark mode during that time.</li>
          <li>Fixed instructions slipping past the protection against prompt injection when invisible characters were hidden inside their words. Removing those characters used to put such an instruction back together unnoticed.</li>
          <li>Upgraded the Visual Briefing assistant (in preview) from the prototype to the beta state. The assistant is now completely implemented and is undergoing a deeper testing phase in preparation for release. To try it, open the app settings, allow preview features down to beta, and then enable the Visual Briefing assistant there.</li>
          <li>Upgraded the vector database behind local RAG (Qdrant Edge) to version 0.8.0.</li>
          <li>Upgraded to .NET v9.0.20</li>
          <li>Upgraded to Rust v1.98.1</li>
        </ul>
      </description>
    </release>
    <release type="stable" version="26.8.2" date="2026-08-31">
      <description>
        <ul>
          <li>Added protection against prompt injection. Documents, web pages, and retrieved content can carry instructions written for the AI rather than for you, for example, text telling it to ignore its rules or to hand over its instructions. AI Studio now always removes such passages before the content reaches a model, while the rest of your document stays intact and usable. When something was removed, AI Studio tells you and can show you which passages it took out. You can turn off the detailed dialog in the app settings. For IT departments: the new setting <code>DataApp.ShowPromptInjectionAlert</code> lets you configure the detailed dialog for your organization. Many thanks to Sabrina <code>Sabrina-devops</code> for implementing this feature and to Simon <code>SimonBpunkt</code> for his work on the detection patterns and their translations.</li>
          <li>Added configurable direct-chat launchers for assistant plugins. Plugin authors and the Assistant Builder can now open a chat with a chosen workspace, provider, profile, chat template, and data sources, while unavailable or unauthorized selections are reported before a chat is created.</li>
          <li>Added the option to load the assistant description from a file in the Assistant Builder.</li>
          <li>Added provider logos throughout AI Studio, making models easier to recognize at a glance. Configuration plugins can now give managed LLM, transcription, and embedding providers their own project icon with the optional <code>IconPath</code> field.</li>
          <li>Added the option for IT departments to enable assistant plugins they rolled out. Approving an assistant only stated that it is safe, so everybody still had to switch it on themselves. An approval can now also enable the assistant, either as a default, which you may switch off again, or in a way your IT department keeps in place. The plugin page and the security card of the assistant tell you which of the two applies.</li>
          <li>Added knowledge about the latest AI models. AI Studio now recognizes Qwen 3.8 Flash, GLM-5.3 Flash, Meta's Muse Glimmer, NVIDIA's Nemotron 3.5, Tencent's Hunyuan Hy3, Grok 4, Claude Opus 5 and Sonnet 5, and Gemini 3.6 and 3.7. It knows what each of them is capable of, so images, videos, tool usage, and reasoning are available right away instead of staying hidden.</li>
          <li>Added the IONOS AI Model Hub as a provider for chats and embeddings. It runs open-source models in Germany, is subject to the GDPR, and IONOS states that your data is not used for training.</li>
          <li>Added LiteLLM as a new LLM provider for chats, embeddings, and speech-to-text. LiteLLM is an AI gateway you run yourself in front of models from many providers. Because your gateway decides where your data goes, you set its trust level yourself. Thanks Prodman Devokadev (<code>prodmanpd</code>) for this first contribution.</li>
          <li>Added speech-to-text for Helmholtz Blablador and GroqCloud, and embeddings for GWDG SAIA. These providers offer these services now, so you can select them when you dictate a message or when you set up a data source.</li>
          <li>Added embeddings and speech-to-text for Hugging Face, so you can now use it to prepare your own documents for retrieval and to dictate your messages. Hugging Face offers both through a few of its inference providers only, which is why you get a shorter list to choose from there than you do for chatting.</li>
          <li>Added a model list for Hugging Face. Until now, you had to type the name of the model yourself and hope you got it right, down to its capitalization. AI Studio now loads the models your chosen inference provider actually offers, so you pick one from a list and cannot end up with a model that the provider does not serve.</li>
          <li>Added more file formats for exporting an AI answer. The export button used to offer Microsoft Word only; it is now a menu which also writes OpenDocument Text for LibreOffice, LaTeX, Markdown, and a webpage. When an answer contains tables, each of them can be saved on its own as a spreadsheet file, named after the heading above it and ready to open in Excel or LibreOffice Calc. This works in the chat and for the results of every assistant. Many thanks to Nils Kruthoff (<code>nilskruthoff</code>) for this contribution.</li>
          <li>Added a choice of file format to the Batch Processing assistant. When it writes one result file per document, those files were always Markdown; you can now pick Microsoft Word, OpenDocument Text, LaTeX, or a webpage instead. For IT departments: the new setting <code>DataBatchProcessing.ResultFileFormat</code> lets you configure the format for your organization.</li>
          <li>Improved the safety of plugin symbols: AI Studio now shows the symbol of a plugin in isolation, so nothing inside a symbol can reach the rest of the app.</li>
          <li>Improved how much memory AI Studio needs. Working with large documents used to grow the app to several gigabytes, and on macOS that memory was never handed back. AI Studio now stays at a fraction of that and returns memory to your system. This matters most on devices with little memory, such as a Raspberry Pi.</li>
          <li>Improved the preview for large documents. It now shows you the beginning of your document instead of loading all of it, so the dialog opens right away. Your complete document still goes to the AI.</li>
          <li>Improved how AI Studio deals with rare internal hiccups. When the app window reloads, or when it briefly loses the connection to its own user interface, work which was still running in the background is now ended properly instead of leaving errors behind.</li>
          <li>Improved how IT departments roll plugins out. A configuration server can deliver any kind of plugin, not only configurations: one archive may carry assistant plugins and further types alongside a configuration, each in its own folder. The folder for staging a test behaves the same way, so a test can mirror the later rollout exactly. The Enterprise IT documentation describes the whole procedure.</li>
          <li>Improved which models you get to choose from when chatting: models you cannot chat with are now hidden. This is most noticeable with a gateway such as LiteLLM, which offers you everything its providers have, including video, live speech, and audio models.</li>
          <li>Changed the model list of GroqCloud. Models which cannot be used for chatting, such as the speech and the safety models, no longer show up among the chat models. The speech models now appear where they belong, in the settings for speech-to-text.</li>
          <li>Changed how plugins your organization rolled out are protected. They can no longer be deleted or edited in AI Studio, which already applied to sharing and replacing them. This also covers plugins staged for a test: such a test now ends by restarting AI Studio or by removing the staged files, instead of through the plugin page.</li>
          <li>Fixed the Hugging Face provider, which had stopped working. Hugging Face changed the way requests are addressed, and AI Studio still used the old way, so chatting failed with a puzzling error about the message format. Chatting works again, and you can now reach far more inference providers, among them Z.ai, Groq, Cohere, DeepInfra, and Baseten. You may also leave the choice to Hugging Face and let it pick the fastest or the cheapest provider for you, which switches to another one when your first choice is unavailable. Should a provider not offer the model you selected, AI Studio now tells you so in plain words instead of reporting a technical problem. The providers Hugging Face no longer run are gone from the list; if you had picked one of them, AI Studio asks you to choose again.</li>
          <li>Fixed the abilities shown for Google's Gemma models. AI Studio did not recognize them at all and treated every one of them as a text-only model, so images, reasoning, and tool usage stayed hidden even though Gemma 4 handles all three.</li>
          <li>Fixed assistants created by the Assistant Builder being named after an internal placeholder, such as "Model decides", when you left the display name empty. The model now picks a fitting name instead.</li>
          <li>Fixed AI Studio reading a document to the end even after you closed its preview. Closing the dialog now stops that work immediately.</li>
          <li>Fixed AI Studio holding on to finished chats, presentation images, and plugin data. It releases them now, so memory no longer grows the longer you keep the app running.</li>
          <li>Fixed assistants handing an outdated result to the chat. When you sent several results without opening the chat in between, you now receive the one you sent last.</li>
          <li>Fixed rare issues when multiple configurations provided introduction texts or mandatory information under the same ID.</li>
          <li>Fixed the description of the global voice recording shortcut always appearing in English. On Linux, your desktop asks you to confirm such a shortcut and shows this description; it now appears in your language.</li>
          <li>Fixed the code editor being offered for assistant plugins your organization manages. Editing one would have withdrawn the approval of your IT department and demanded a fresh security audit.</li>
          <li>Fixed assistant plugins your organization deployed alongside a configuration not being recognized as centrally managed unless they declared it themselves.</li>
          <li>Fixed a misleading warning in the log when an organization deployed an archive that carries no configuration of its own.</li>
          <li>Fixed AI Studio underrating what many models can do. Newer Claude, Gemini, Grok, DeepSeek, and Qwen models were missing abilities they actually have, such as image input or tool usage. This was most noticeable with OpenRouter, where nearly every model was affected. AI Studio now derives these abilities from the same source for all providers, so a model offers the same capabilities no matter which way you reach it.</li>
          <li>Fixed the abilities shown for the models offered by Mistral. Mistral names its models after their release date, so almost every one of them was missing image input or reasoning: picking Mistral Large from the list gave you a different set of abilities than picking the very same model by its full name. AI Studio now goes by the release date and gets all of them right, including the Ministral models and the open-source models Mistral hosts, such as GLM.</li>
          <li>Fixed AI Studio staying silent about why dictating or embedding failed. When a provider explains what went wrong, you now get to read it instead of a general note that something did not work. If a provider cannot handle the audio format AI Studio sends, it says so and suggests contacting that provider.</li>
          <li>Upgraded Rust to v1.98.0</li>
        </ul>
      </description>
    </release>
    <release type="stable" version="26.8.1" date="2026-08-19">
      <description>
        <ul>
          <li>Added Hetzner's experimental inference API as an LLM provider. It runs open-source models in the EU and supports text and image chats through its OpenAI-compatible API.</li>
          <li>Added support for the new open source models DeepSeek V4 Flash and Pro, GLM 5.2, Kimi K2.7 Code and K3, as well as Qwen 3.6 and 3.8.</li>
          <li>Added a prototype Visual Briefing Assistant that turns documents, data, images, audio, and video into self-contained interactive HTML briefings. When you want to test it, you have to enable this preview feature in your app settings.</li>
          <li>Added organization-configurable defaults and visibility controls for the Visual Briefing Assistant.</li>
          <li>Added a share button for assistants, configurations, and language plugins. It uses the native share dialog on Windows and macOS. For Linux, we added an export option, which stores the plugin archive at a location of your choice. When you work on a translation for a new language, you can now hand your current state to testers or to us with one click.</li>
          <li>Added the option to install plugin archives from your files: use the import button on the plugin page or simply drop an archive onto that page. Assistants, configurations, and language plugins are supported, and plugin archives now have their own file extension <code>.mwplugin</code>. Before installing a configuration, AI Studio shows what it sets up: which LLM providers and data sources it adds and where each of them sends your data, plus how many settings it takes control of. A configuration takes effect right away and has no on/off switch, so please install one only when you trust its source. You can remove it again at any time.</li>
          <li>Added a delete button for assistants, configurations, and language plugins you installed or placed yourself. Until now, such a plugin could only be removed from the data directory by hand, which was especially painful for configurations because they have no on/off switch. Before deleting a configuration, AI Studio lists what disappears with it, such as providers, data sources, and settings that return to their default. When you delete the language plugin you had chosen, AI Studio returns to choosing your language automatically. Plugins shipped with AI Studio and plugins deployed by your IT department cannot be deleted.</li>
          <li>Added options for organizations to disable importing, sharing, and exporting plugins, with a separate option for configuration plugins. Organizations can now let people import assistants while keeping configurations to their IT department.</li>
          <li>Added a priority for configuration plugins. Organizations that deploy several configurations can now decide which one wins: a configuration with a higher priority overrides the settings and providers of a lower one. This allows a company-wide base configuration that each department refines for itself.</li>
          <li>Added the Batch Processing Assistant: process all documents of a folder in one run. Each document is sent to the AI along with your instructions – either a free prompt, one of your document analysis policies, or instructions you import from a file. You choose whether each answer is stored as its own Markdown file or whether all answers are collected in one CSV results table, which you can name yourself. Every run writes a log that lists each document with its processing time, the model, the status, and the reason for any error. When you start another run on the same output folder, we ask whether you want to continue it: documents that failed or are missing from the log are processed again, which is helpful after a crash or when documents exceeded the context window of your model. A single failing document never stops the run, and you can cancel at any time. The assistant was contributed by Jens Erler (<code>j-erler</code>) and marks his first contribution to AI Studio. Thank you, Jens, for this wonderful and useful contribution.</li>
          <li>Added a way for IT departments to try out a configuration before rolling it out. A configuration placed in the new <code>.config-tests</code> directory below the plugins directory acts like one your organization deployed, including the approval of assistant plugins, so a test shows exactly what colleagues will see later. No configuration server is needed for this. AI Studio empties that directory every time it starts, so a test configuration is valid for one session, and the information page reports it while it is active. The Enterprise IT documentation describes the whole procedure.</li>
          <li>Added providers for which you bring your own API key. Until now, a provider your organization configured had to come with a shared API key, which meant your IT department needed one key for everybody. Such a provider can now be handed out without a key, so that everyone signs in with their own, for example, a personal OpenAI or Anthropic account. This works for chat, embedding, and transcription providers alike. The provider stays managed by your organization: the host, the model, the instance name, and everything else remain fixed, and the only thing you can edit is the API key. In the settings, these providers carry a key icon instead of the usual lock, so you can see at a glance where you have to add your key; for embedding providers, the test button stays available, so you can check your key right after entering it. Your key is stored on your device in the operating system's credential store, and it stays there even when your organization withdraws the provider later, so it is still in place should the same provider return. For IT departments: the new <code>AllowUserProvidedAPIKey</code> option does this, and it works for <code>LLM_PROVIDERS</code>, <code>EMBEDDING_PROVIDERS</code>, and <code>TRANSCRIPTION_PROVIDERS</code> alike.</li>
          <li>Added CSV and TSV files to the file types you can attach. AI Studio was already able to read them, but they could not be selected.</li>
          <li>Improved how your organization's configuration behaves when a configuration plugin is present but cannot be loaded, e.g. because of an error in the plugin. Such a plugin still manages your app, so its settings, providers, data sources, profiles, and chat templates now stay in place instead of being removed.</li>
          <li>Improved reading large files from slow locations such as network drives. AI Studio now waits considerably longer before it gives up, and it tells you when it does.</li>
          <li>Improved how Word documents (<code>.docx</code>) and OpenDocument files (<code>.odt</code>) are read. AI Studio now reads them itself instead of handing them to Pandoc, so these documents no longer need a Pandoc installation. It reads them section by section, which keeps even large documents responsive, and it now picks up more of the document: the title, the author, headers and footers, footnotes, endnotes, and comments. This was contributed by Nils Kruthoff (<code>nilskruthoff</code>), who also wrote the library behind it. Thank you, Nils, for this great contribution.</li>
          <li>Improved the order of your models. Every list of models is now sorted by provider and name, so all models of one provider stay together. Until now, models appeared in the order they were created, which meant that a model added later always showed up at the end of the list. This was especially confusing when your organization rolled out new models. In the settings, the tables for LLM providers, embeddings, and transcription are now grouped by provider and start collapsed, so even a long list stays easy to survey.</li>
          <li>Improved how you access log files from the information page: you can now open their locations in the system file manager or jump directly to the Log Viewer Assistant.</li>
          <li>Changed how approvals for assistant plugins combine when your organization deploys several configurations. They now add up, so a department can approve additional assistant plugins without repeating the approvals of the company-wide configuration. Previously, the last configuration replaced all earlier approvals, which silently required a new security check for those assistants.</li>
          <li>Fixed attached files reaching the AI as empty documents when AI Studio could not read them. The AI then answered as if your file had no content, and nothing pointed to a problem. AI Studio now names the cause instead, for example, an unavailable network drive, a file another program is blocking, a protected PDF, or a scanned PDF without a text layer, and it no longer attaches such a file.</li>
          <li>Fixed files that are open in another program being reported as an unrecognized file type. AI Studio now tells you that the file is currently open elsewhere and asks you to close it. This also works for files on shared network drives, where a colleague might have the file open.</li>
          <li>Fixed files with a wrong file extension being reported as empty. AI Studio now recognizes what a file really is by looking at its content, for example, a PowerPoint presentation that was renamed to <code>.txt</code>, and reads it accordingly. It also points out the wrong extension, so you can correct it.</li>
          <li>Fixed files whose content is not text being sent as an empty document. AI Studio now tells you that the file is not readable as text, which usually means it carries a wrong file extension.</li>
          <li>Fixed executable programs with a harmless file extension being read as text. They are now recognized by their content and refused.</li>
          <li>Fixed a single unreadable page of a PDF silently cutting off the rest of the document. The remaining pages are now used, and AI Studio tells you which pages are missing.</li>
          <li>Fixed a single unreadable sheet of a spreadsheet silently dropping all remaining sheets.</li>
          <li>Fixed PDFs, text files, spreadsheets, and presentations requiring Pandoc. Only HTML files need Pandoc now, so every other file can be attached and read without it.</li>
          <li>Fixed attached files that are temporarily unavailable, disappearing from your message without a word. This could happen when a file was stored on a network drive.</li>
          <li>Fixed the file preview showing an empty document when reading the file failed. It now shows what went wrong, so the preview again answers what AI Studio will hand to the AI.</li>
          <li>Fixed the file preview looking like an empty file while AI Studio was still reading it. Larger documents and PDFs need a moment to be read, and until now that moment looked like a file without any content. The preview now says that it is still loading and shows the content as soon as it is ready.</li>
          <li>Fixed problems while reading files being missing from the log file after the first one. This made exactly those issues hard to track down that only appeared later on.</li>
          <li>Fixed reset buttons in assistants. As you may have noticed in the Document Analysis Assistant, resetting it could leave content from the previous analysis visible. Reset buttons now clear previous results completely.</li>
          <li>Fixed dropping files after you closed a dialog that accepts files itself. Such a dialog takes over dropped files while it is open, but never handed that role back when you closed it. Afterward, the chat and the assistants silently ignored dropped files until you switched to another page. Each time you opened such a dialog again, the problem got worse.</li>
          <li>Fixed configuration-managed settings, remaining active after their configuration plugin was removed.</li>
          <li>Fixed settings not returning to your own value after a configuration was removed. When a configuration takes control of a setting, AI Studio now remembers the value you had chosen before and hands it back once no configuration manages that setting anymore. This covers an IT department withdrawing a configuration, deleting one yourself, and an administrator ending a test configuration. When a configuration only suggested a value, and you changed it afterward, your choice stays as it is.</li>
          <li>Fixed the integrated code editor to keep errors and other issues in plugin code visible in the footer while scrolling.</li>
          <li>Fixed the trusted badge so you can now see at a glance which models are trusted. It is shown consistently for self-hosted models and models from trusted providers.</li>
          <li>Fixed approvals for assistant plugins being accepted from any configuration plugin. An approval marks an assistant as safe without a security check, and the app states that your organization approved it. Only configurations your IT department deploys, or that an administrator stages for a test, can do that now; approvals from any other locally placed configuration plugin are ignored and reported in the log.</li>
          <li>Fixed withdrawing a configuration your organization deployed. A configuration that declared itself as locally managed stayed on the device even after the IT department stopped deploying it, and it kept every right of an organization configuration, such as approving assistant plugins. Where a configuration is stored now decides this instead of what the configuration says about itself, so withdrawing one always takes effect. This also applies to a device that was offline while the organization changed its policy: the withdrawal is applied when AI Studio starts again.</li>
          <li>Fixed preview features contributed by several configuration plugins at once. Only the most recent contribution was recognized as coming from your organization, so features enabled by another configuration looked as if you had switched them on yourself. Each configuration is now tracked separately, which lets your organization enable one preview feature company-wide and another one for a single department.</li>
          <li>Fixed which configuration wins when two configuration plugins collide, e.g. by claiming the same plugin ID, by managing the same setting, or by defining the same provider. Previously, this was down to chance, so a local configuration plugin could take over parts of the configuration your IT department deployed. Configurations from your organization now always win, and every ignored attempt is reported in the log.</li>
          <li>Fixed the assistant categories when your organization hides individual assistants. A category heading could stay visible above an empty area, and the Log Viewer could disappear together with the Localization assistant. Each heading now follows the assistants actually shown below it.</li>
          <li>Fixed a removed API key staying in the operating system's credential store. When you cleared the API key of a provider, the previous key remained stored and was still used. It is now removed together with your change.</li>
          <li>Fixed AI Studio installing a second copy of itself next to an existing installation. Its updater always installs into your personal user folder, so an installation elsewhere, such as one your IT department rolled out, was never replaced. AI Studio now recognizes those installations and leaves them alone. The information page tells you which case applies to yours. For IT departments: automatic updates can now stay enabled for everybody, and the Enterprise IT documentation explains the rest.</li>
          <li>Fixed installing an assistant on Linux when AI Studio runs as a Flatpak. The Assistant Builder was able to create an assistant, but installing it always ended with an unexpected error.</li>
          <li>Fixed the plugins page and the assistants page reloading again and again on Linux, which started as soon as an assistant was installed. Both pages kept flickering, and nothing on them could be used anymore until you left for another page.</li>
          <li>Removed the legacy PowerPoint format (<code>.ppt</code>) from the selectable file types. AI Studio has no reader for it, so such a file could be attached but never read. The modern <code>.pptx</code> format is not affected.</li>
          <li>Upgraded dependencies to their latest versions to improve security and stability.</li>
        </ul>
      </description>
    </release>
    <release type="stable" version="26.7.3" date="2026-07-21">
      <description>
        <ul>
          <li>Added support for OpenAI GPT-5.6 Sol, Terra, and Luna; Anthropic Claude Fable 5 and Mythos 5; and Google Gemini 3 Flash, Gemini 3.1 Flash-Lite, Gemini 3.1 Pro, and Gemini 3.5 Flash.</li>
          <li>Added support for OpenDocument presentations (<code>.odp</code>) when attaching and reading presentation files.</li>
          <li>Added a log viewer assistant that shows AI Studio log files in a read-only view with search, log filters, highlighting, and auto-refresh.</li>
          <li>Added audio and video transcription for chats and assistants. AI Studio now prepares supported media locally, sends only normalized audio to the configured transcription provider, and attaches the resulting transcript instead of the original media.</li>
          <li>Added AI-assisted editing and revision for assistants created with the Assistant Builder. Thanks, Nils Kruthoff (<code>nilskruthoff</code>), for this contribution.</li>
          <li>Added options to view and edit the code of AI-generated assistants and to delete your own generated assistants. Thanks, Nils Kruthoff (<code>nilskruthoff</code>), for this contribution.</li>
          <li>Added enterprise configuration options to hide the last changelog and vision panels on the welcome page. Thanks, Dominic Neuburg (<code>donework</code>), for the contribution.</li>
          <li>Improved the "My Tasks Assistant": you can now provide one or more documents in addition to text or use documents alone when asking to identify tasks.</li>
          <li>Improved presentation imports so AI Studio can include speaker notes, slide comments, and presentation metadata in the extracted content.</li>
          <li>Improved update guidance for Flatpak installations and added an enterprise option that lets organizations manage updates entirely through their IT department.</li>
          <li>Improved secure API-key storage diagnostics on Linux. AI Studio now provides specific guidance when the default password collection is missing or locked, a password-manager prompt is dismissed, or no compatible Secret Service is available.</li>
          <li>Improved the file dialogs to prevent opening multiple times when you click "Open" or "Save" multiple times in a row.</li>
          <li>Improved assistant plugins so they clearly indicate when content from a document has been loaded and clear the indicator when the assistant is reset.</li>
          <li>Improved the Assistant Builder security check so it can use the selected provider when no dedicated security audit agent provider is configured.</li>
          <li>Fixed an issue that could leave AI Studio unresponsive after waking the computer from sleep. Yes, we know this was an annoying bug, and we apologize for the inconvenience.</li>
          <li>Fixed connections to internal HTTPS services and enterprise configuration servers that use organization-provided root certificates on Linux.</li>
          <li>Fixed enterprise configuration plugins from Windows-created ZIP files may not load correctly on Linux when the ZIP contained plugin files inside a folder.</li>
          <li>Fixed the voice recording shortcut on Linux so it works globally on supported desktops and while AI Studio is focused on other Linux desktops.</li>
          <li>Fixed voice recording and transcription on Linux.</li>
          <li>Fixed copied content from AI Studio not remaining available on the clipboard on Linux.</li>
          <li>Fixed dragging and dropping files from the home folder into the Linux Flatpak version.</li>
          <li>Fixed being able to switch document analysis policies while an analysis or media transcription was still in progress.</li>
          <li>Fixed file extension handling so files are recognized correctly regardless of uppercase or lowercase letters in their extensions. Thanks, Paul Schweiß, for reporting this issue.</li>
          <li>Fixed AI Studio failing to start on Linux systems &amp; showing an outdated version on the Flatpak page.</li>
          <li>Upgraded Rust to v1.97.1.</li>
          <li>Upgraded .NET to v9.0.18.</li>
          <li>Upgraded Tauri to v2.11.5.</li>
          <li>Upgraded common dependencies.</li>
          <li>Upgraded runtime dependencies.</li>
        </ul>
      </description>
    </release>
  </releases>
</component>