Examine the contents/context of the image and save the standardalized data to Sheet

I created a quick AI, which is expected to do a job, scan the image and extract all the text. And then standardalized data, such as DATE, ISSUER, AMOUNT OF MONEY, etc. Bit of miss-readings for some of the documents, but generally working fine. (although processing time is bit too long…)

We added a task to save the standarlized data onto Google Sheet. As a result, AI generated a sheet and saved to Google Drive, but the problem is the format was in MS Excel rather than Google Sheet, and also the Excel was created number of time, i.e. per image/document. I expected the AI will save the tabular data to a single spreadsheet. Is it possible to instruct AI to do so?

1 Like

Let me try and get back to you. At the moment, we have implemented a few file formats including CSV file (which I suspect looks like Excel). But there are many more to add. We should be able to do this rapidly.


On my screen (for Google Drive), the file type was indicated as Excel, so it is sure the AI genereated Excel instead of CSV.

It is better to have optional control if or not AI would generate SINGLE Google Sheet to store the generated data or make multiple ones based on the users use cases.

Hi Koichi, we have a simple export of the whole collection as a UI action (doesn’t need an LLM for this). That is hidden at the moment but we’ll try to expose this by early next week.
Also, this is possible for the LLM as well in case it should be done automatically (we’ll get a sample up and running to demonstrate that soon)

1 Like

Just added the export/download capability in the UI. This emits a CSV file.

If you add tasks “foreach” row, then you are telling your agent to do that once for every row. That is probably why it is creating one file per row.

If possible, please share the thunk you were using so we can look at it. I’d like to see why you were getting something in Excel format. Thanks

How to share a thunk with someone else? …
Go to the Team icon in the left control pane of the thunk and add the email address of the person you want to share it with. In this case, please share with me: praveen@thunkai.com

Hi Praveen, I just followed the guide and added your private email to share my AIs.

Thank you. Yes, I can access it.
It looks like something odd is going on there.
Each row is supposed to get its own task.
Will investigate and get back to you

Thanks for looking into it and waiting for your futher feedback.

Hi Koichi, we have found a bug exposed by your thunk. Could you please also add scott@thunkai.com as a team member so that my colleague Scott can debug. Should be easy to fix once we can run it through a debugger. Thanks again for your patience.


Hi Praveen,
I added Scott. Done.

1 Like

Thank you. We have fixed the problem you ran into. My suggestion is to abandon that old thunk and create a new one.

I created an example for you and added you to it to see what I did. Please LMK if it is helpful. In all thunks, you will have an initial task that is used to “collect” the list of items or tasks. And then each item will have its own task or sequence of tasks to process it. In each task, the AI agent operates in the context of the particular related data item and the instructions of that particular step of the plan

Hi Praveen

I created new AI agent from the scratch, assuming the agent would do : -

1 Extract text from image.
2 Get the standarlized data, such as Date, Total Amount, issuer.
3 And then save such a standalized/extracted data to the spreadsheet as tabular format

However, I see task 1 and 2 have been managed, but 3 can not be accomodate. You can see my conversation throught the executed task panel to see how I was confused and struggle.

In the meantime, by chance, I see the spreadsheet AI agent created (not sure how I did…) but once we save the number with thousand separator, AI agent take comma as column separator. Bit of a bug.


You should not need the AI agent to export it to the sheet. There are a couple of things to be aware of:

  1. Unlike AppSheet, in Thunk.AI, the data storage is native to the platform. So even if you import data from a sheet initially, the Thunk has the current data. And it updates the data as the agent operates, so there’s no need to tell the agent to write data.
  2. If you go to the Data pane, you will always see the latest data.
  3. You might need to export data occasionally. If so, there is a UI control to do this (see the “Download” icon at the bottom of the Data pane).
  4. If you really want your AI agent to export data, you can go to the “Collect” task associated with the list/collection and ask it to write out the current data set to a spreadsheet

Does this makes sense — in brief, the Thunk data is always uptodate and you can choose to export/download to a CSV/spreadsheet when needed

Export/download could be inevitable options we should have, but more importantly, how to access to data seamlessly from external sources.(Ideally, for instance from AppSheet , we can access/crud the Thunk AI data/table as perfect solution.

In the meantime, I opened up “Data Outline” however I could not find any extracted data, and UX table stay blurry.

See how the data outline tab looks like now.

Or do I miss something? Is there any other place I should access to see data?

Step status to save data seems to be “not started” due to some reason? is this the reason why I could not see any data as table view?

Sorry for questions in chain, but I just wanted to make sure. I managed to “add step” to the plan details and instructed to save data to Google Sheet directed by file ID. Step was created, but this type of step would not work with Thunk AI?

Is my understanding correct?

ok, I understand the confusion!
What you are looking at here is the Planning pane. The planning includes the process flow and the data structure. What you are seeing is the data structure (aka “Data Outline”) which is part of the plan.
But when the thunk is running, you want the Data pane (first icon on the left)! In your thunk, you will see four entries, I think.

On the same item, only one task is run at a time. So while the"Extract text from image" is still running, the next task will not start.
When you open the current running step, at the bottom of the chat window, you’ll see a “Finish Task” button in yellow/orange. Once you click that, it tells the AI agent that it is ok to move on. And that will automatically start the next task in the process.

The AI agent has a bunch of capabilities. It can only do things within those capabilities. In the next week, we will provide that list to see, and also to add to. It does, for example, have the ability to write a file iin Google Drive. But it does not currently have the ability to update one row in a spreadsheet, or to edit one sentence in a Google Doc, etc. It cannot read your Gmail, but you can forward individual emails to the thunk.
Also, btw, it does not even have access to your full Google Drive from a permissions point of view. You can optionally provide that permission, but we don’t ask for it by default. So by default, it can only access files that you select via the file picker.

So in short, the AI agent is limited by the capabilities we have provided it or have permissioned it for. This set of things will grow over time. This week, we are adding the ability to invoke a webhook, so that the AI agent can send things to external systems. And we are also adding the ability to listen to a webhook so that it can receive things from external systems. Suddenly, that will open up a lot of possibilities via integration with other things like Zapier that can connect many other systems to webhooks.