We’ve released “AI Workspace Data Expert Assistant” chatbot where you can ask questions about your Fibery data.
It is experimental, so you can enable it in Settings → Experimental Lab.
This post is to share your questions and answers. Do share anything useful, unexpected, interesting, or completely broken.
Unfortunately we always get the same message
Is there a way to select the databases that the AI searches in?
It seems you have quite large databases schema. We are thinking how to make it work for such cases. Thank you for the feedback!
Here is my feedback:
The AI Workspace Data Expert is a great improvement, and offers instant mini-research results on fibery workspace data. I would fully put the development power on that, and make it work well together with other Key features like Highlights.
There are many things to discuss and ideate around this, but all of them are and will likely be depending on one feature that is not in place:
Because my primary need is to connect information, I miss that the output of the AI Assistant and the AI Workspace Data Expert cannot be saved as entities automatically. It requires manual copy pasting.
Why? Because most of the time the output needs to be further processed, combined, evaluated, worked with, either through automations or other processes.
For example, storing the output as reference, being able to highlight and relate to other content.
I treat AI input as similar to user input or meeting input. There is no logical reason why to treat it differently from a procedural perspective.
Workspace Data Expert Assistants are merely ‘an AI capability’.
Next logical develpment step is to allow Fibery AI Agents to have similar funtionality as users, but without the same user accounts (however, with AI agent accounts). Agents can be assigned to tasks, meetings, and engage in discussions.
I also am unable to get a single response, all result in for example:
400 This model’s maximum context length is 16385 tokens. However, your messages resulted in 98413 tokens. Please reduce the length of the messages.
Unfortunately in your case there is no chance it will work. Your schema is way too large for any AI assistant. Maybe we will add more narrow context, like per-Space, and then it will have a chance
Thats a good idea I think, because specific information is often part of one or two spaces. I think a selection of spaces would be good since an organizational process and thus data often is spread over more than one space.
I’m just starting to experiment more with the AI assistant.
Is there a way we can give context to the AI assistant? Some naming conventions or logic may be industry specific or company specific. For example, if I write out some context in the database description or a field description, does that help give context?
I think per-space would be an improvement. And not just if a workspace is too big, sometimes context is hard. Two examples:
In addition to per-Space, I actually would love if you went even narrower to per-Entity. I want to be able to ask a question about an entity, with the AI assistant auto filtered to just that entity and its relationships.
A lot of entities wouldn’t need this, but I think this would be extremely useful for “big entities”, the ones with the most relationships (for me it would be project pages, client pages, vendor pages). Just clicking 
icon in the top right of that entity.
I really like the idea of context-dependent AI, if it could be implemented intuitively, alongside broader AI querying/'assistant" capabilities. The example of a “magic wand” button on each entity by the other controls/functions seems like a good place to start.
