Since the beginning, Tana was always meant to work with AI. After extensively testing with our users on the possibilities, we are now officially building and integrating AI into Tana. We're calling this Tana AI.
Tana AI provides access to the best LLMs in the world. We pair each task with the ideal model for the job. Every model will already be fine-tuned for use in Tana, so you don't have to worry about that.
All savings we receive on volume pricing are yours. When you subscribe to Tana Core, you get access to not only all the fine-tuned AI models we subscribe to, but also the volume pricing we receive.
There is no need to bring your own AI API keys to Tana anymore. But if you want to use your own key for some things, we have left two commands (Ask AI + Generic AI query) that you can use with your own OpenAI key.
AI should be like electricity: you turn it on, and it just works.
There are many features in Tana that work with AI. For an extensive list, see the next section on Features using AI.
You can chat with your knowledge using AI in Tana! AI chat comes in two flavours: Ask AI where you get a standard chat, and Ask AI about this content which takes into consideration the context you add into the chat.
Talk to Tana and get your thoughts transcribed straight where you want it. Use live transcribe or audio-enabled fields and tags to speak straight into Tana.
AI actions require credits to run. Tana Core users get 5000 AI credits every month. Credits can be topped up.
AI is a Tana Core feature. You can only use AI in Tana if you are on the Tana Core plan.
You can create a New thread based on the conversation up until that point, if you want to take the conversation in a new direction. This puts the full chat up to that point into the chat context.
Hover over an AI response, and you can Insert the latest response to the parent of the chat (the place you initiated the chat from), Copy the response to your clipboard, or continue chatting to it.
You can Stop/rerun an AI job if you want to tweak the prompt. This will delete the AI response up to that point. Hold Alt+click to create a branched path of conversation that keeps the history of the previous response(s).
For more tips on how to use AI chat, check out the AI chat guide by pressing the i (info) button in the chat.
Events: We have added a new way to trigger commands to happen when nodes are added or removed from another node. They are called Events, and can be set up for any node and supertag.
For nodes: Run Configure node and go to On child added/On child removed:
For supertags: Go to Supertag config > AI and Commands > On added/on removed
The command line Open GPT log monitor will open a monitor that shows you how many credits are being used for various AI work. Right now it only shows credits spent using OpenAI models. We hope to add support for other models in the future.
Click on the line items to expand and see the payloads.
The Tana Labs experiment AI for builders is what kickstarted the work on AI in Tana over a year ago. All the things you were able to do with AI for builders, you can do with a Tana Core subscription, without the need to bring your own key.
InfoWe have removed the Cluster nodes command, which was an early experiment of what was possible with AI and embeddings. We might bring back a better/more robust version in the future. ()
ImprovedAsk AI via the command line now uses GPT-4o, and a lot of places where we were using GTP3.5 by default have been upgraded to GPT4o-mini, which is smarter, faster and cheaper. We are no longer using GPT3.5 by default anywhere, but it is still available for manual configuration in an Ask AI command. We have therefore removed the GPT4 switch from the UI. ()
ImprovedWe've made some changes to AI prompts and models for field suggestions, suggest table columns, AI fields, autocomplete etc which should give more stable results. ()
ImprovedThere is now a more robust uploading of AI generated images, and we show a visual loading indicator for AI generated images in banner and image nodes while images are being uploaded to our server. ()
FixedFor certain AI models, the system was ignoring users request to use their own OpenAI API keys.
This 🔒 has been fixed. Your keys will now fit 🔑 ()
FixedWe fixed validation of OpenAI keys to support longer keys. (OpenAI API sk-proj key input is now accepted) ()
ImprovedAdded support for GPT-4o-mini when choosing model ()
ImprovedUpdated latest GPT-4 implementation to AI fields ()
ImprovedIn "Manage your account", we now show the monthly reset date for AI credits, and dates now show actual dates, not relative dates. ()
How fast does Tana update to the latest AI models?
Sep 12, 2024
The team strives to make the models available as soon as possible. New AI models require fine-tuning before they can be used in Tana. This involves adjusting the prompts that ask AI to deliver information in Tana format. Additionally there are many out of the box AI features that need to be tested for consistency.
The turnaround for this depends on engineering capacity, usually 4-8 weeks. We can make new models available in Ask AI and Generic AI commands much faster, so we often stagger the rollout.
Does the GPT log monitor show AI credits usage by the Meeting agent?
Sep 12, 2024
Only partially; the transcription of the meeting does not use OpenAI and is therefore not logged in the GPT log monitor.
Meetings use a service called Soniox and we are working with them to expose the AI credit cost of individual meetings in a simple way, but currently it's not possible.
The post-processing of the meeting transcript is however done using OpenAI and will be visible in the GPT log monitor.
If you have Tana Core, you can use the 5000 AI credits that are part of the plan, along with your own Open AI API key restricted to the Ask AI and Generic AI Query command nodes. AI Chat and Meeting agent will always use the Tana AI credits.
Use AI to connect Readwise tags to my list of topics
Sep 12, 2024
The Tana API that is used to power the Readwise integration is exactly the same (and therefore has the same limitations) as the Input API. It can only write new nodes to Tana, and the API can only write a single value to new fields.
So, to get all your tags out of Readwise, we recommend a comma-separated format, and then using AI to link to the correct nodes in your graph.
Here's a video of how it works (no audio):
Here is the prompt:
PROMPT
Look at each keyword separated by the commas:
Input: [${rwTags}]
And try to find equivalents in this list of all keywords: //Insert search node here that finds all keywords/topics