Iām excited to share that Ask AI is now officially available to all Softr users ā including on the Free plan.
This is a major milestone for us. Ask AI is more than just a feature - itās a built-in teammate that lets your app users ask questions and get instant answers from their data. No setup complexity. No risk of the wrong people seeing the wrong data.
A few key highlights:
Secure by default
Ask AI only pulls data that your users already have access to - nothing more. It fully respects your appās visibility rules, user groups, and permissions, ensuring every response is safe and tailored to that user.
How to enable it:
Go to any dynamic block (thatās powered by our DSV2 datasources)
Navigate to Actions ā Add Ask AI as a topbar button
Airtable, HubSpot, and SQL support are coming in August
Note: Ask AI is currently block-level only, with full app-wide chat coming next
Free & unlimited - for now
All plans (even Free) get unlimited Ask AI credits for a limited time. Eventually, AI usage will count toward monthly credits depending on plan level.
How you can support us:
Try it out in your own apps and explore its power
Share your experience on LinkedIn, Twitter, or in Slack
Help us spread the word by liking, reposting, or commenting on our launch announcement.
Weāve been working hard to make sure this launch is not just exciting, but responsible and aligned with our vision of safe, powerful app-building. We hope you all enjoy it!
Great feature!! Eager to use it with airtable (I almost only have databases on airtable with scripts). I was able to test the voice function from the computer but on the phone despite the microphone activated, it doesnāt do anything, am I the only one in this case?
Looks like a great feature, just tried it on a a few blocks.
The design and the integration is very smooth fits right in. Puts AI feature availability front and center on everyblock. The predesigned question to prompt a chat were well thought out.
I love the feature.
However, The information is limited to dynamic blocks and the information stored in that particular table only. I guess good for people with strict access requirements.
Any plans to convert it into a App wide Chat bubble? (Where access requirements are not so strict?). I think I saw it mentioned somewhere, an ETA would be nice.
Use case:
I have a LMS with almost 100 lessons at this points. Different dynamic blocks for different chapters and section modules.
Since it is a LMS, the access requirements for a user are not so strict or needed at all for me. The use Case is more geared towards a better integrated and a well trained AI model.
I do not have any sensitive content. So I would love to train this AI on my entire Database and make it available as a genuine AI assistant with side wide available and knowledge.
My Concern:
ā I know (as the founder) the Ask AI feature only have access to a particular source/Table.
ā But my users donāt know about this limitation. They will assume AI should know everything inside the LMS and should be able to solve and answer basic questions even if the answer is stored in a different table. When they try and AI is not able to answer a basic question from a different module stored in a different dynamic block/Table, they will assume AI is dumb and not trained well.
At the moment I am using Botpress with a Chat bubble embedded on the whole App. I am training my Open AI Assistant on the whole Database table by table. So this Botpress chat-bubble embed AI knows a lot more that just the information in 1 dynamic block.
But of course the inbuilt design and intergration of softr AI is lot better and sleeker.
While your team is working on Adding it as a Block for App wide Content, I am thinking of 2 ways to make it App wide already, just need your input:
PS: I am thinking of Doing all this within Softr Databases.
Is this AI currently reading content from the docs attached in the respective blocks or Rows?
I am thinking, ā> What If, I Make a simple 30-40 Page doc with all my LMS content and teaching material and Add it in every Row of the source table. This way if the AI is able to read the doc then it will have knowledge to app wide content and lessons.
If 1st is not possible, Second way I am thinking of doing is to just put all the content of the 30-40 page doc into a long text field. But I know you guys recently added some limits here but I can work around them by adding multiple long text columns I guess. This Data is just for AI knowledge base not for display, so i guessing Long text field appraoch should also work.
Do you think, either of these 2 ways would work to increase the overall knowledge of the Ask AI function ?
I will of-course try it today or tomorrow. But just asking if you have already thought of this way to quickly make the AI feature App wide?
It may be helpful to update the help docs and add this as a suggestion for other users as well.
Can I ask, before I offer this to my users (weāre still in testing), is there ANY chance that youāll restrict it to Professional plan and above only? Were on a basic plan. I donāt want to get users used to it only to find you then take it away.
This will feature will remain available to everyone on a paid plan, including basic.
However, it will be on a credit system in the future. Where you can purchase credits for your users to use the feature. More details will come in the coming weeks for this.
This is a great feature but can I ask a few questions about the prompt that pops up when you enable it please?
Enable āAsk AIā action?
Let your users ask questions about data in this block via an AI bot. All filters and user permissions applied to the data will be respected.
Please note that:
*By enabling this feature, you agree to send your data to 3rd-party AI model providers.
*This feature is currently in beta. Usage limits may apply.
Iād like to understand the privacy conditions that āAsk AIā is currently subject to. Ideally, weād need it to match our ChatGPT Team planās privacy settings as follows:
No Training on Your Data
OpenAI does not use conversations, file uploads, or any other data from ChatGPT Team accounts to train its models. This includes all interactions within your workspace, whether through chat, file uploads, or code execution.
Private to Your Workspace
Everything you and your teammates do within the ChatGPT Team workspace is private and accessible only to members of your team and OpenAI systems strictly for operational purposes (e.g., system maintenance, abuse detection). It is never shared outside your organisation or made public.
Data Retention for Support Purposes Only
Data may be stored temporarily to provide the service and support functionality (e.g., viewing your chat history), but again, itās not used for training and is not visible to OpenAI employees except in very rare, highly controlled cases such as abuse investigations, and even then only when necessary.
@Jjenglert Just been testing this out and it is awesome.
Howeverā¦
There are massive issues with data privacyā¦
e.g. When I test as a user from one business and ask it about another businesses details (with no relationship to the test user) then the AI tells me everything I want to know about their confidential data (inventory, orders, revenue etcā¦)
Can we apply global data restrictions to this? How do we trust this to be tailored to each user as it is certainly not āSecure by defaultā and can definitely not be using this as violates all policiesā¦
Hey @Grainne
I would be careful.
We have removed the Ask AI from our app as it was releasing confidential information to users about other users and their business.
I have reported to Softr but we will definitely be leaving this in test mode until Softr can clearly demonstrate it working securely for us.
This is a bit surprising, I found the security to be pretty good and somewhat restricting for my usecase.
I had the other problem, I want the AI to have a wider knowledge base because I have a LMS. The AI only had access to limited information if the AI setting was - Configured fields only. This made the AI feature pretty dumb and restricted.
I had to change the setting to - Any fields from Source
Then this AI had access to the knowledge in entire source table and all the course material. Made it a little smarter. But still it was limited to 1 Source Table only. So I had to consolidate all my courses in 1 table. Now I have an AI trained on all my courses and able to help answer questions.
I have not tried this on a user specific data table, Not needed in my use case.
But you may want to play around the AI setting. If it is set to anything from source then it will have access to all the data in a particular table, but limited to only that table. I have not seen any spill over or data leakage case.
But may be i am just missing some test case, would love to know more.
Ask AI is quite secure, it just depends on how you configure it like all things.
make sure youāre using conditional filters to secure it further.
and global data restrictions do work with it.
So make sure youāre setting it up to your liking. If you still see problems, let us know. Most likely, it will be related to your database structure and the ways your structuring your rules.
Yes all our global data restrictions and further conditional filters are applied. We have no issues with the data vis for users.
This has only come up with the Ask AI when kept probing questions about other businesses which sometimes it wouldnt give data about but then sometimes it wouldā¦
Is there a chance the Ask AI could vary responses between Preview Mode and Published Mode?
@FPX indeed in preview mode it could re-use some of the history in chatā¦ but thatās for you as admin previewing multiple usersā¦ we will fix this but it should not be a concern outisdeā¦
e.g. open two browsers different browsers and ask the same questions you will see no data will come along
This was the very first chat we used, so it cannot have been caused by using chat history.
I have tested Live mode vs Preview mode and can confirm there are variances in responses to the same questions.
I have also done further testing and found more issues with data privacy.
As per the screenshots below:
Blue = the User asking questions to Ask AI
Yellow = another userās business (the User in blue does not belong to this business, but is a customer of it)
In this example, Blue asks what the revenue of Yellowās business is.
Ask AI responds by providing the total revenue for Yellowās business, not restricted to only Blueās orders.
This means Ask AI is currently not enforcing global data restrictions and is pulling sensitive business data outside the userās scope.
Thanks @FPX for the update. We wonāt be activating it for testing purposes until we know more about what data is being collected.
@Jjenglert@artur do you have any information/documentation on this please?
Weād like to understand the privacy conditions that āAsk AIā is currently subject to? I donāt think its enough to say āAsk AI is quite secureā
@FPX you can go into studio click on app settingsā¦ then publish/preview and check your queries again ? we made sure that as a builder when you test/preview multiple users from the same browser still no history is being shared
@Grainne when configuring the AskAI you configure if all fields or only blockās used fields are available to be analysedā¦ there is not more than thatā¦ and the questions you ask and their answers are not used for training anything else
āBy enabling this feature, you agree to send your data to 3rd-party AI model providers.ā
So what data is it sending and where is it sending it to?
Ideally, weād like access to documentation around this feature, specifically:
Which third-party AI providers are used
What categories of data are sent
Where the data is processed
How long itās retained
Whether itās used for model training
Can we disable the feature at any time without data remaining in the AI providerās possession?
Without clear answers, we canāt assess the risk of data exposure, leaks, or misuse, which means we canāt use the feature (and weād really like to use it )
