Jan 11, 2024 / Chatbots (experimental), Filter by Rich Text snippets in Views, shorter trial & no free version

[mdubakov]

Chatbots (very experimental)

The Fibery Chatbots allow users to ask questions and receive answers from an AI-powered chatbot right in Fibery. The chatbots use information from specified Fibery databases, so it can be a useful tool for finding information quickly.

Chatbots is an experimental feature, enable it in Settings → Experimental Lab.

You can create several chatbots that use several databases to answer questions or do other things. For example, we’re trying Product Expert chatbot that knows all Fibery Features and Insights.

Chatbots work nice if you have some knowledge base and want to find an answer to specific question. For example:

• You store User Guide of your product in Fibery. Here you can try questions like “How to create a formula?” or “Do you have Linear integration?”
• Knowledge base about competitors, here you can try questions like “Does X support feature Z?”
• Some domain-specific knowledge base, like Content Marketing tips and tricks.

Check Chatbots User Guide. We are eager to learn from you and see what use cases you will invent with Fibery chatbots. Please do share your feedback!

This feature is a result of Slow December .

Row Height in Table View

Graduated from experimental. We’ve moved Row Height setting into Rows, fixed all found bugs and improved wrapping for all columns.

In Table View header use field names always instead of unit name

We’ve made a little tweak to help keep your tables neat and tidy. From now on, we’re ditching the extra words in column names, and sticking with just the original field name. So, “Avatars of Assignees” will simply be “Assignees”, “Progress (progress bar)” becomes “Progress”, you get the idea.

Rich-text snippets: filter can be applied in Views

Rich-text Snippet is supported in the filter, sort, and colors. It can be used in all views. The most frequently requested use case is to filter out entities with empty rich test field, now you can do it.

And it works in the Automation as well:

Rich-text snippets: character cut by card size

The snippet size is based on the card size now.

No free Solo plan and 14-days trial

We’ve removed free Solo plan and reduced trial from 30 to 14 days. Let’s see how it goes.

• All existing accounts are not affected, this change only affect new accounts.
• If you use Fibery for free as a solo user now, please enjoy it, we don’t have any plans to restrict that.

Fixed Bugs

• Table View fixes:
  • Last row gets unchecked when selection is done via shift+
  • Copy-paste is overriding the last row if paste was happen to the “new row” placeholder
  • Selection get cleared for some items when user select some single items and then array of rows using +shift
  • Wrong selection array when user selects rows via shift +
  • ‘Add new item’ does now work in table view when last visible row isn’t 1st level one
• Newly created entity doesn’t disappear from relation list after unlink without page refresh
• Small layout issue in integrations
• Formula with IF(Condition, DateRange, DateTimeRange) does not work
• Separator lines are almost invisible in ‘light there’+‘dark menu’ mode
• Search does not work in Inbox filters

[YvetteLans]

Wow, what a release It’s insane how much you guys can do in a week / month / not-so-slow-december

Will have a good look at all the new functionalities and provide feedback afterwards.

@mdubakov

In Table View header use field names always instead of unit name

This is so awesome It makes grid views a lot cleaner.

Currently when you combine multiple databases in a grid view, you will have two columns with the same name if

• One is a formula (text)
• And one is a rich text field (snippet)

Is that something that can be fixed?

Else I need to turn off the snippet since I mostly use formulas. Not a real big deal but it’s such a beautiful feature

[mdubakov]

@YvetteLans

Currently when you combine multiple databases in a grid view, you will have two columns with the same name if

• One is a formula (text)
• And one is a rich text field (snippet)

Technically this is relatively hard to implement, so I am not sure we will go for it in near future.

[aklavrov]

I’m very excited about this release. Thank you for your work

[YvetteLans]

@mdubakov

Technically this is relatively hard to implement

That was what I expected Will hide it for now.

[njyo]

Wow, so many goodies! You release more things than what I can find time to play around and experiment with!

Happy 2024!

[mdubakov]

This is nice to hear. We will try to keep this pace

[Illusory]

@mdubakov

Do you have Linear integration?

Now you’re just teasing!

[Yuri_BC]

Thank you for this great release, looks awesome.
One comment:
Row Height in Table View - when shown as field view in an entity, when selecting Medium/Tall/ExtraTall, the field view height does not expand according to its content. The result is that you always see only one entity (partly) listed, unless you scroll inside the embed window. It would be better to show the same amount of entities, but make the field view vertically longer to accomodate reading its content for larger row theights.

[Yuri_BC]

@mdubakov

Chatbots (very experimental)

This to me is the most exciting and helpful new feature. Thank you!!

I tried it out, and as mentioned it needs to be polished, since …
Local fibery account data:
“I’m sorry for any confusion, but as an AI, I don’t have the ability to access or locate information in your Fibery data. My main function is to provide information and answer questions based on the context provided to me. I respect user privacy and confidentiality.” (when mentioning Fibery data)
LLM (intenet) data
“I’m sorry, but as an AI, I don’t have the ability to access real-time information or updates from the internet. Therefore, I can’t provide the historical information about who the president of the United States was in the year 2000. My responses are based solely on the context provided to me.”

[mdubakov]

@Yuri_BC

Local fibery account data:
“I’m sorry for any confusion, but as an AI, I don’t have the ability to access or locate information in your Fibery data. My main function is to provide information and answer questions based on the context provided to me. I respect user privacy and confidentiality.” (when mentioning Fibery data)

Did you try to index some exact database and ask questions related to it?

[Yuri_BC]

@mdubakov

Did you try to index some exact database and ask questions related to it?

Yes I did, and it does generally give relevant answers, but the AI doesn’t seem to know its using Fibery data, so any mentioning of Fibery makes it deny it knows anything (unless I have an entity that explains how Fibery AI chatbots work, then its part of its privided context).

Somethig else…

Clearer distinction between references and AI search results

‘See Also’ AI search results:
Each answer ends with a line See also: with two to three links to fibery resources, but the AI is unaware of this listing. Apparently you included an AI search to the answer, but that list confused me because I thought that these were references of fibery resources that the AI used to formulate its answer.
References used in formulating the answer:
When asked, it can list resources it used from its provided context, but these are not the same as the “See Also” listing. So this may need to be clarified to users. Also, the references that the AI can list upon request are plain text titles of the resources, and would ideally be clickable links.

Suggestion:
So basically it would be very helpful if an anwer has two separate listings (may want to allow users to turn then on or off):

1. References: The answer includes top 3 resources as clickable links that the AI actually is aware it used to formulate its answer.
2. AI Search Results: The ‘See Also’ footer starts for example with “Separate AI search results (not references of the answer)” to inform the user.

[dukevannori]

is there a way to train the chatbots on documents? i have quite an extensive knowledge base as files in a folder, but i only can select databases … or am I missing the setting? (or is there a “best practice” that is NOT putting data in files and folders?)

also: does “bots are exposed to ALL fibery users” mean, that its unrestricted to all users of MY workspace … or ALL users in general … so all all? I hope you mean the first …

[Oshyan]

Chatbot is an exciting new feature with very high potential! I agree it has some significant issues at present, though my experience has so far been different from @Yuri_BC . For me it seems like there are somewhat random clickable links scattered in the responses, probably derived from in-entity links, but not directly relevant to my query. It seems that the info sources are often directly referenced in the response but only in non-clickable text, and then those same referenced entities are often linked at the bottom. Seems like the in-line (in the response) references to source material should be clickable and the seemingly irrelevant included links should be removed/re-evaluated.

The following issues may just be related to incomplete indexing, pending further update and testing… My bigger concern, though, is lack of clarity in what is being indexed and how. I am getting coherent responses that reference my data, but it seems to have an extremely sparse sampling of that data. It pulls out very atypical examples when asked general questions about the data. To put it in the context of a practical product-oriented use case, it would be like indexing the documentation for a product with a Connection Manager (and an entire page/entity about this feature), but when asked about how to use the Connections Manager, it cherry-picks ~3 examples of where “Connections Manager” is referenced/linked to in some other feature, rather than directly referencing and linking to the full page that documents the Connections Manager itself. Kind of odd.

Maybe this is just due to partial indexing of the content so far, in which case an “indexing progress” indicator somewhere would be good (perhaps there is one that I haven’t found yet). But I did see responses that more clearly indicated insufficient/no content had been indexed and this is different…
Ah, I now see the little colored “light” indicator with on-hover status indication. Nice, glad that’s already there! A progress indicator/percentage would be a lot more helpful if possible (also now a bit concerned with API token use, unsure how much this will use up! Hah).

Another question in the meantime: can the chatbot work on indexing already done previously for the AI Search?

[Yuri_BC]

Some other Chatbot needs:

1. Make available the complete AI Chatbot threads history as entities
2. Allow whole Chatbot Chat thread conversion to any other entity
3. Allow Chatbot single Chat message to convert to entity
4. Allow Chatbot in sidebar of each entity, just like the AI Assistant (or merge those features)

[mdubakov]

@dukevannori

is there a way to train the chatbots on documents?

Not yet

@dukevannori

also: does “bots are exposed to ALL fibery users” mean, that its unrestricted to all users of MY workspace … or ALL users in general … so all all? I hope you mean the first …

Your workspace users for sure. Eventually we will add an option to “publish” chatbot, but so far it is not added

[dukevannori]

so the only way to train the bots on data is to put every document in a database description field? is there a way to convert documents to database entries?

[Oshyan]

After 24hrs both of my added Chatbots still show “Reading”/yellow light status. I may have tried to create too many of them close together (I made 3 over the course of an hour or so), not realizing that indexing took longer than I thought. But it does seem like things are stuck here since since my OpenAI key usage has not changed since yesterday. And unlike the AI Search Indexing there is no way to reindex a chatbot as far as I can see, you have to just remove and recreate it, which is unfortunate. So far promising but buggy. I remain excited to test this further but will hold off until there is more info on indexing issues, etc.

[pzol]

@mdubakov can we have variable height rows? Can the row height setting serve as a maximum height instead of always taking up the full height??