The new AI Chat is useful, but there are a few problems. First of which is that you can’t recover deleted chats from trash.
IF we treated AI chat like a field (kind of like comments are now), you could make a “AI Chat” database, and someone could go in, make a new chat, and talk to the AI. Where it could really shine is that multiple people could open the same AI Chat entity and see live answers at the same time. Everyone sees the chat and its treated as an entity. Ideally it would also know about what entity it sits in. So you could also as an AI Chat field to a projects database, then in each project you can ask it for a project update, and it will know what project its in. This would integrate the AI further into fibery ecosystem.
Then when entities are deleted and restored, the ai chat that was there with it gets restored too. The only down side is that its a to-one. The field gives you one chat, which is the same for all. If you want to start a new chat thats not really possible.
A solution would be to relate a project to chats (one to many), then you can see all the chats that project has. The only to make the UX as good would be to introduce “One at a time” multi-relation view. Where is shows an embedded entity with its fields (AI Chat) and a entity switcher to switch to a different one from the relation.
Maybe maybe maybe. But ai chat for a single entity (like project) could be useful? Open a project, then ask ai questions about it. Then the turn output to entity button can be a button to append / overwrite to a rich text field in that specific entity.
So while I see that the entities wont be so useful as the output, there is still a reason for keeping the old chats in an AI chat (the fact is that everyone does it). Doing it in the way outlines above would just allow it to be better integrated into the Fibery ecosystem.
If there was an “AI Chat” rich-text field type, then a Chat that happens “in” this field (in a particular entity) could supply the entity as context to the Chat, as well as supplying the complete, cumulative Chat history.
So, a handy way to:
Define a context for a specific Chat, anchored to an entity;
Remember the Chat history and include it as context for future requests made from that same field.
This is interesting direction, but Field primitive does not look very promising to me. It looks like too complex thing to be a “Field”. It is close to Comments linked to database entity. Anyway, context setting is a thing we want to solve for AI Chat, so it is very relevant problem.
Agreed, and I gave some more thought as to what an AI agent really is. A field is a container of data specific to an entity. Even "Call" or "conference" field type in Fibery under the hood would be a separate auto generated channel link per entity.
An AI agent is more of an function than anything else. Takes an input, maybe does something, then returns an output. This input and output is knowledge and data, but the function itself is not. So I agree that maybe this idea was suggested too quick. Note for not so far future, getting the full output in automation steps like so: Do automation actions return step results? Can we use them further down the automation stack? - #4 by gb_carbone, would allow for lots of flexibility to build your own AI Workflows without the need for so many automations. Right now we can’t create a new linked thing, then update it’s description with the smart agent’s response. It needs to be two seperate automations, one to create, and one triggered on creation.
The only thing, @mdubakov, is that I haven’t been able to use the Smart Agent in rules… It works in button presses, but i can’t get it to work in any way in rules. Not sure if Im setting it up wrong or its a bug still.
When it gets fixed, I have an idea to make my own “AI Chat” using just automation actions. I’ll send a video here when it works. Then it knows context, and can be given permission like any other entity in fibery. The only thing I’m confused about is that it might query and return data the user doesn’t have access to… i’ll play with it and report back!
I am actually now building an MAS (Multi Agent System) in Fibery and a lot of its operational functionality is not done by AI at all, but by automations.
This has actually benefits if done well - because it gives the reliability needed which AI does not have yet, and likely may not have very soon.
The idea is that in fibery I create an AI Language that can be used in automation triggers.
Its alsocalled structured prompting or schema-based prompt engineering, which refers to designing prompts in a consistent, machine-readable format so AI outputs can be reliably parsed and turned into actions or workflows. This approach is often combined with function calling or domain-specific prompt languages to automate follow-ups like entity creation or task execution.
So the Smart Agent that Fibery now is building could do a lot, but I think that the management and control of the chained Smart Agent calls is more interesting for Fibery users.
It returns a value to “what can you do?”, but has authentication problems… I can’t get it to authenticate and run a call. I thought it was a problem with the fact that it doesn’t know what user triggered it, but I tested it and it does show the right user who triggered.
Example output:
AI Chat Count Information
I’m unable to directly query the number of AI chats in this workspace due to authentication limitations.
To find out how many AI chats exist in your workspace, you can:
Navigate to the AI demo space in your Fibery interface
Go to the AI Chat database view
Look at the total count typically displayed at the bottom of the list view
Alternatively, you can create a Board or Table view of AI Chats which will show the total count in the interface.
This information isn’t accessible through the current conversation context alone, as it requires querying the entire database, which needs proper authentication.
Would you like me to help with something else related to your AI chats?
The potential you just tell it what chat its related to and can then query the chat for older messages and respond the desired message.
The big question is authentication. For things that are not triggered by a user, what data should it have access to? Maybe this can be set by the person making the automation?
You can do lots with this!!
I think this makes more sense that AI as a field. I will rename the title.