Great idea, we will think
Claude Cowork and Fibery has the potential to become an absolute power couple
2 Likes
From claude - could you allow more room so information isnt lost?
One thing to flag: the full document had to be slightly condensed in a few of the supporting sections (Values, Vision & Chemistry, some OKR detail) to get it through the Fibery API in one write. All the substance is intact but you may want to glance through it to make sure the tone still feels right in those areas.
Could you improve back linking from entity text? It seems to help AI if there are more backlinks between referenced things in the entity text.
Eg i have a guide entity that references the employee benefits guide; could you create a simple backlink automatically?
It would help grow the web of information and keep everything nicely related for AI to follow.
Can you also allow claude to create backlinks in text blocks using # like a human can; it cant seem to do it at the moment.
(I tried to get fibery AI to go through and create backlinks in rich text and it failed)
EDIT: this is actually really important to solve to make claude cowork and fibery work better together - it is causing issues
I have been able to get Claude to create entity mentions ("chips) with a custom MCP we built, was this what you referring to?
So if I have a task “Take out the trash” in a task db, Claude can find it and mention it if I comment in another entity. So say I have a db of meetings, and after a team meeting today I come back and comment in the meeting, that “oh by the way I completed ‘take out the trash’” today and I forgot to tell the team in the meeting…I could in my set up tell Claude to write that comment and tag the task and it will do that, with a full regular #mention.
Now you can see how each change was made in Fibery August 13, 2026 / Custom Apps, See how each change was made (UI, API, automation, AI)
1 Like
The discussion above about organizational LLM knowledge bases and a possible Fibery Architect Skill connects closely to something I’ve been running into while using Fibery with external coding agents.
I think there may be one additional layer between “the agent has the right knowledge” and “the agent can safely reason from that knowledge.”
I’m using Fibery as a semantic/read-model over several external systems, with things like claims, evidence, decisions, provenance, supersession and unresolved questions represented in Fibery while the original systems remain authoritative for the facts they own.
What surprised me is that:
an agent can retrieve exactly the right records and still reach the wrong conclusion.
A Skill can teach the agent how to query and reason, and better retrieval can give it the right context, but some epistemic state seems to need to persist in the model itself.
Three examples have been particularly important:
not queried ≠ absent ≠ unavailable
If an external authority was never checked, the absence of a record cannot mean “nothing exists.” Likewise, a failed lookup cannot become evidence of absence. I eventually had to model acquisition/check state explicitly.
corroboration ≠ authority
Several systems can report compatible information while having different roles for different questions. A repository, execution tracker, package registry and correlated Fibery record might all agree — but for “what package is actually published?”, one of those has a different epistemic role from the others.
So the missing relation isn’t just:
record → related record
but something more like:
question → expected authority → observation → conclusion
AI synthesis ≠ independent evidence
If an agent reads several records, creates a useful synthesis and writes it back into Fibery, the next agent needs to know that this is a derived statement — otherwise AI output can eventually become apparent corroboration for itself.
I’ve ended up modeling this at the application layer with concepts roughly like:
Claim → typed Evidence → expected authority checks → historical evaluation → contradiction/provenance
and giving agents structured context reads rather than expecting each new conversation to reconstruct those semantics from search results.
This may be exactly the kind of domain model Fibery expects users to build, rather than something Fibery should own.
But as Fibery moves toward being an organizational knowledge layer for external agents, I’m curious where you see that boundary.
Do provenance of source-vs-derived knowledge, authority/precedence, acquisition state, and unresolved contradiction remain entirely user-defined domain concepts — or do you imagine Fibery eventually having some agent-facing primitives/conventions for this kind of epistemic metadata?
I’m also curious whether you’ve seen other teams hit this once their Fibery workspace becomes persistent memory for multiple agents.
Happy to share a concrete failure case and the model I ended up with if useful. And equally happy to discover that I’ve built around a Fibery-native primitive I’ve missed.
2 Likes