ImprovedUpdated default prompt for Text Processing Agent to achieve better summary results for most meetings. We recommend tweaking your summary prompt for better results for specific meeting types
FixedAI prompts would sometimes not care to wait for the search results to update before sending them. Also 'Refresh search' command line now also wait, so any dependent command nodes running after it will have a updated search to reference.
NewThe Tana Template Store (early version) is now live! You can browse and install a growing collection of official Tana templates, directly inside the app.
To check it out, click Supertags in the sidebar and then “Browse templates.”
NewCORRECTION: There will soon be a way to reset your primary workspace! Go to your home node options > Reset workspace. Instructions and screenshots in the Workspace doc.
InfoWe will now extract action items with the text processing agent as long as there is a "Action item target". If you want them tagged as well, configure the "Tags to use for action items" as well
ImprovedSome cleanup of text processing agent parameters: removed some deprecated ones and reorganized the most important ones up top. No change for already configured commands.
InfoAlphabetic sorting is based on a more intuitive expectation now:
(A b a A B) would previously sort to (A A B a b), whereas now it will sort to (A A a B b).
ImprovedWe now automatically include Source materials in AI chat as context (you can remove the node from the chat message, if you don't want it included)
ImprovedPasting Markdown (with correct mime-type) now at least *tries* to read the headers and make an outline from that. It also supports some **markdown** _formatting_ now.
NewNew command node: Open node. Specify which node to open (current node, current day page, Library, Schema, Inbox, Source node, or any reference), and which panel position to open it in (current or new panel). Use the context node to specify which workspace the relative node lives in.
ImprovedUpdated service we use to provide Live transcriptions to a new version. It offers better quality transcripts, handles multilingual conversations better, and also covers more languages than before.
Live transcription
ImprovedWe’ve added a new default Auto option to Live transcriptions, which should work well in most cases. If you find that it hesitates between similar languages (Swedish for Norwegian for example), you can choose a language. English will always be added as an additional option no matter which language you choose.
Live transcription
ImprovedAction items and recognized entities produced by the text processing agent are now inserted as tag suggestions, giving the user better control of whether they are relevant or not. We also stop showing two suggestions if a node with a tag suggestion is inside a field with options from tag.
ImprovedNew command line for all tags with base tag meeting: "Start meeting with #tagname", creates a new meeting node and starts live transcription immediately
ImprovedEnabled cmd+click on command nodes in menus (compact menu etc) to open command definition in side panel (this was already working for normal command buttons)
FixedFixed some cases in the Command Line where custom commands triggered by a keyboard shortcut wouldn't launch if they overlapped in name with existing commands
InfoChange in meeting processing prompt to not add links back to transcript, as part of cleanup. We might bring this back in a better version in the future.
InfoAdvanced/hacker feature, comes with no guarantees: A node or supertag's Icon (that replaces the node bullet) can now take a Field definition, and it will use the field value as its avatar. Now; if any value put in the Icon field has its own icon, Tana will use the icon instead of the text in the status.
ImprovedInvoking @-mention, tagging and field search inside a draft chat will now look up entities in the workspace that the chat is in, not the user's private workspace. New nodes, tags or fields are also created in the workspace the chat lives in.
FixedSmall fix for Rewrite commands - we no longer include the "Voice memo captured" title which sometimes led LLM to return results in English rather than the target language.
InfoWe received complaints about some nodes with a Checkbox field==No were not showing up in searches. The reason is that a new checkbox field is initially in a "Not defined" state when added to an existing supertag. Until we have a better way to solve this, there are now two intermediate solutions to this: 1. Run "Update default values" on the checkbox field in tag config, or 2: Use the new "Not defined" Checkbox data type field option in the query builder, so these undefined fields can be found too.
ImprovedNew command line: "Show today for this workspace" to quickly go to the Today node of workspace you are in. That workspace needs to have a 'Daily notes' calendar node already.
FixedFixed bug that sometimes led commands with Target node inserting nodes into a live search. Note that if you want multiple nodes to be inserted as children of the target node, you also need to set Insert strategy: as a child.
FixedOur tooltips were showing up below the context menus. Now fixed.
FixedThe tag config overcounted how heavy a tag would be when a field inside the tag had many values. This showed a warning that was not intended to be shown (since the real node count was often many many times lower).
FixedFixed issue where prompt workbench would sometimes not open from a field, and made sure default model is the same as AI fields use (currently gpt-4o).
NewWe’ve introduced new error messages for our Event system. The Event system is what automatically triggers commands based on actions like "On added/removed" or "On child added/removed." As our command system has become more advanced, it’s easier to create loops or recursive behaviors that may run indefinitely. If one of these behaviors is detected, a fail-safe will trigger, and an error message will appear. To reset the Event system, simply refresh Tana. We’d love to hear from the community about any commands that regularly trigger the fail-safes, especially if you believe your commands are working as intended. Thanks! —@Fei
FixedStop using timezone for date objects where there is no time specified. Today = today, tomorrow = tomorrow, no matter where you are in the world. Thanks Darren for your patience on this, more fixes to come!
FixedFixed commands that wouldn't run when 1. they invoked a command line action that required expanding such as "Move to", and 2. had custom shortcuts mapped to them.
InfoCommands nested under other commands will no longer be available in command line. If you want it to be available in command line, put the actual command in the graph somewhere, and reference it as a sub-command.
ImprovedWe now support o3-mini in most AI commands, such as Ask AI and AI Chat. We currently only offer "reasoning effort" set to the default, which is medium. This has improved cost per token over GPT-4o, which about the same performance. However, reasoning tokens might cause costs to increase in total.
ImprovedWe've improved pasting of lists. We've had several edge-cases related to ordering and indentation that should now work much better.
If you are still experiencing paste results that look bad it would be very helpful if your bug report includes a reproducible case. The easiest way for us to fix it, is it you take your example and go to https://evercoder.github.io/clipboard-inspector/ and paste it in there, then copy the information from "text/plain" and "text/html".
FixedWe made a change to AI prompt processing to automatically expand all search nodes, but this caused far too much irrelevant content to be sent to AI processing in some cases. We are reversing this change for now, note that you can always specify that reference nodes should be expanded by using the Expand references system field.