- Added your mailboxes (e-mails) as a new kind of data source, as a prototype. AI Studio keeps a local index of your mails, including the ones you sent and your drafts, and the AI can search, read, and count them. It works with every mail provider that offers IMAP, among them Gmail, GMX, WEB.DE, iCloud Mail, and an Exchange server of your organization with IMAP switched on. Microsoft 365 and Outlook.com follow later because they require a sign-in with Microsoft. To try it, open the app settings, allow preview features down to prototype, and enable both RAG and mailboxes.
- Added the number of indexed mails to the information page, next to the number of indexed files.
- Added tool collections for tools that only make sense together. Search Mails, Read Mail, and Count Mails appear as one entry, Mailboxes: you select them together, they require one confidence level, and IT departments switch them off together.
- Added organization-wide management for mailboxes. IT departments can prevent users from adding mailboxes, and they can rule out the less strict choices of where a chat may send data after reading mails.
- Added the mail servers of your organization to the setup of a mailbox. IT departments can name them in a configuration, and the mailbox dialog lists them before the public mail providers, with a hint on the username and a link to their instructions.
- Added a way for IT departments to allow mailboxes only on the mail servers of your organization. AI Studio then connects to no other server, and the AI no longer reads a mailbox somebody added elsewhere.
- Added a setting to share which features of AI Studio you use with the operators of self-hosted AI servers, usually your organization. When you switch it on, requests to self-hosted providers and LiteLLM name the feature that sent them, such as the chat, an assistant, or an agent, together with the version of AI Studio and your operating system. Cloud providers never receive these details, and AI Studio sends nothing to its developers. The setting is off by default, and you find it in the app settings. Thanks, Dominic Neuburg (`donework`), for this contribution.
- Added a way for IT departments to let requests to the self-hosted AI servers of your organization name the feature of AI Studio that sent them, either for everybody or as a default that users may change. This shows which features are used. The Enterprise IT documentation explains what is sent and what to clarify before you switch it on, for example, with your works council.
- Improved the Read Web Page tool. When you have not allowed the AI to choose web addresses freely, the tool now refuses every address that does not appear in your chat or in what the tools returned, instead of only asking the AI not to read it.
- Improved the installation on macOS: AI Studio is now signed and checked by Apple, so it opens like any other app. You no longer need the Terminal.
- Improved the installation on Windows: AI Studio is now signed, so Windows shows who published it. For a while, Windows may still ask you to confirm the installation until it knows the app better.
- Improved the question whether AI Studio should transcribe your audio and video files. It now waits until you choose Yes or No, so a click next to it no longer drops your files.
- Improved how fast AI Studio starts when the configuration of your organization includes tokenizers. AI Studio now checks such a tokenizer only when it has changed.
- Improved the button in the chat for moving the chat to a workspace. You can now use it whenever the workspace list lets you move the chat as well, for example, right after you attached an audio or video file and before you sent your first message.
- Improved how AI Studio keeps the message you have not sent yet. When AI Studio stores your chats automatically, your message and its attachments now stay with the chat, and they are back when you open the chat again, for example, after you switched to another chat.
- Improved how AI Studio keeps the message of a new chat. When AI Studio stores your chats automatically and you leave a new chat before you send your message, AI Studio now stores the chat with your message. You find it among your disappearing chats, or in its workspace when you started it there.
- Improved how AI Studio indexes large local data sources. It now works through them in steps of a few minutes, so your other data sources, such as your mailboxes, stay up to date in between.
- Fixed a warning about the confidence level that stayed in the dialog for adding a local data source, even after you chose an embedding provider that meets it. The dialog now checks again whenever you change either of them, so a warning also appears when you switch to a provider that falls short.
- Fixed two confidence levels sharing the same name in the German user interface. Moderate is called "Mäßig" again, and Medium stays "Mittel".
- Fixed the Search Confluence tool missing pages that the search of your Confluence wiki finds. The tool now searches the same way as the search box of your wiki.
- Fixed AI Studio starting only partly when the configuration of your organization took a while to apply. The plugins late in line never started, so the app stayed in English instead of your language, and the configuration of your organization was missing, together with its models and its welcome text. Now every plugin gets its own time to start, and a slow one no longer holds up the others.
- Fixed AI Studio deleting the providers, chat templates, and profiles of your organization, including their API keys, when the configuration of your organization did not start in time. They now stay in place until the configuration starts again.
- Fixed saving the settings taking about 20 seconds each time when a chat template had an attachment on a network drive that was out of reach. Since AI Studio saves its settings while starting, this also slowed down every start.
- Fixed local data sources taking up more and more space on your disk with every change to your documents. AI Studio now removes what the vector database no longer needs.
- Fixed answers breaking off with an error message from self-hosted AI servers such as vLLM when the AI made up the name of a tool or sent broken details for one. AI Studio now tells the AI that its tool call was invalid, and the AI can finish its answer.
- Fixed OpenAI models using about 4,300 more tokens than necessary with every request. AI Studio offered them OpenAI's own web search in the background, even when you had not selected a web search. To let a model search the web, select the Web Search tool below the message field.
- Fixed AI Studio sometimes rebuilding the index of a data source from scratch when the configuration of your organization was applied while the data source was being updated.
- Fixed AI Studio is continuing to update the index of a data source with the old settings after you, or the configuration of your organization, changed its embedding provider or the expert settings of the data source, which could take hours for nothing. AI Studio now stops right away and starts over with the new settings.
- Fixed AI Studio is continuing to send the content of a data source to an embedding provider after you deleted that provider, or after a confidence level changed so that the provider may no longer read the data source. AI Studio now stops right away.
- Fixed data sources are staying unavailable after the configuration of your organization changed their embedding provider, until one of their documents changed, or you restarted AI Studio. When AI Studio updates your data sources automatically, it now rebuilds their index right away.
- Fixed justified texts being hyphenated by the rules of English even when AI Studio shows another language, such as German. Screen readers now also know which language AI Studio uses.
- Fixed AI Studio pausing for a moment while it opened the local index of a data source, such as a mailbox. Other work, for example applying the configuration of your organization, had to wait until then.
- Fixed a security issue in how AI Studio prepares images for the Visual Briefing assistant. AI Studio now protects all parts of its internal connection in one central place, and an automated test checks that no part is left out.
- Fixed the transcripts of your audio and video files disappearing when you edited your last message in a chat. You no longer need to attach the files again. Thanks, Bernhard, for reporting this issue.
- Fixed the message you had not sent yet losing its text or the transcripts of your audio and video files when you moved the chat to a workspace.
- Fixed AI Studio sometimes warning about unsaved changes when you left a chat in which you had only selected a chat template, without writing anything yourself.
- Fixed regenerating an answer clearing the message you had started to write.
- Fixed the warning about unsaved changes in a chat appearing only by chance when you save your chats manually or have turned off workspaces. AI Studio now reliably asks before you open another chat, create a new one, or leave the chat page, so you no longer lose your changes by accident.
