August was productive month with quite good focus. Initiatives assessment:
Granular Access. Core is still the bottleneck.
Growth. Good progress.- Fibery Apps. Release and many follow up features.
- Make AI Agents happy. Good progress, many things released and some more started.
- Other. In general OK, many small things here and there, but it is expected.
Granular Access
Allow to hide Space data from Admins is moving forward with the majority of potential critical vulnerabilities closed.
Do not allow Admins to access private data in Spaces via scripted Rules looks like the final boss, unless we uncover more.
The first release in early October (optimistically) looks possible.
Oops
Tracking all permission changes in history was a great idea to troubleshoot “why Alex doesn’t have access anymore”, lower the support load for us and anxiety for architects.
This is especially noticeable when some event turns out to be missing and three people spend hours trying to solve a puzzle: 
Event about revoked author (Created By) access is missing from history. Fixed now!
Also, Missing notification about assignment when user gets access to the entity via People field is something we haven’t considered properly before — up until now.
Growth
It’s frustrating when you use all of your trial AI credits, purchase a Pro subscription…and get nothing. So we’re going to Bump monthly usage limits on the first purchase.
A couple of edge cases:
- WS should be charged for extra usage after subscription is cancelled
- “Billing to” section is missing on invoices
Fibery Apps
Official release of Custom Apps did happen in August and 1600+ apps were created by users already. Custom Apps usage quickly catch up with Timelines and Report views, will they overcome Calendar and Dashboard views in a month?
Weekly trend for new apps creation is relatively stable, in total we see ~2K iterations for all apps per week (iteration = prompt execution AI agent):
More and more users create custom apps via MCP and they iterate more heavily when MCP is available (why not, it is way cheaper!) So it seems it was a good move to allow Custom Apps creation via MCP.
Overall, Custom Apps is a very successful feature and we will expand it deeper.
Done
- Visual edit → click element, describe changes
- Version history for Apps
- Duplicate App action
- New Embed → Generate App inside rich text block
- Edit code manually in App builder
Next
We want to bring more functionality into custom apps, and this requires app platform update. So now we are working on this mechanism and then will focus on more functionality in custom apps, including better rich text component, opening entity view in panel, allow to render automation buttons and auto-update app UI when data changes in the background.
- Invent a flow for “platform” update for Custom Apps
- Increase App chances to survive schema changes (renames, etc)
- Open Entity View in panel from custom app
- Switch on some lib for querying, like tanstack/query
- Better rich text component for Custom Apps
- Allow custom apps to render Automation Buttons
- Add routes so exact page in app has own url
We will research 
Custom AI build Entity View and maybe even start building a solution.
Make AI Agents happy
Finally we started to track MCP usage and it is more popular than internal Fibery AI:
- We have ~1400 users using MCP every week, while only ~500 use Fibery AI.
- 540 active workspaces use MCP, and 380 use Fibery AI.
It means MCP usage is growing fast and our bet on MCP was right. People tend to stick to Claude and ChatGPT and do not rely on AI inside other tools much.
It does not mean we should not have internal AI Agent, but it should not be pushed to users
Done
Quite many things were released. Now it is easy to identify who made a change, undo massive changes and create rules and buttons via FIbery AI or MCP
- Track additional info about how the change was made (MCP, API, etc)
- Show Details for changes in automations rules history
- Views history (activity log for Views)
- Add “Via” indicator for comments created not from UI
- Batch undo/undelete after mass accidental changes
- Fibery AI and MCP: Update small part of rich text and add comments
- Create an automation (Rule or Button) with AI (attempt #2)
- Create an automation rule or button via MCP
Next
History improvements are still coming, but the end is near:
- Integrate document history into entity history almost ready, waiting for unlock from core
Rich text improvements are also there:
- Suggestion mode for documents We aim for October, but :kostik: is very pessimistic about this estimate.
Fibery AI Agent is in re-work mode:
- Migrate Fibery AI agent AGAIN (to JITT) is in finalization stage. Still work to do, but should be released in September. It should make Fibery AI internal agent better, more stable and allow us to provide all kinds of models with it
More tools for Fibery AI and MCP and maybe even better Schema API:
- Access data from whiteboard in MCP and Fibery AI
- Create & Update Whiteboard View with AI so far experiments are not great, so hope is fading out, but we are still trying.
- Better Schema API some chances to start implementation in September.
We are also working on AI Integration Portal. It will be easier to architects to create custom integrations soon.
Other
There will be always other things, here is what was done and what we are going to do next.
Done
- Convert Field: single-select → multi-select
- Convert field: select → Database
- Slack: fetch reply threads via separate requests
- Desktop
- Auto-load tabs content when opened in background
- Close tabs with scroll click in desktop app
- Hide Required Fields (with value) on Forms
- Currency exchange: allow to select what exchange pairs to sync
Next
Some assorted, but useful things will come:
- Watch entity rich edit field change Since we will have unified history of rich text field and entity, it is natural to watch changes in rich text fields.
- Share (aka export) Space as template: include Files and Images
- [Desktop] Add support for multi-window mode
- Show affected Formulas, Lookups, Views, Rules, and Buttons when deleting a field or just curious
- There is a chance that we will explore better search options to support filters for every database, but with degraded precision for such cases with only keyword search via “contains” operator. In a nutshell, we will give people an option to create temporary views during the search fast, maybe it will aleviate the problem with structural search
- Customize load limit (aka page size) for Relation View (Table, List)
- Filters: When lower level is filtered, higher levels must be filtered correspondingly by default (hide empty lanes)
- Pinned Filter: Filters inside Pinned Filters (and sort)
- If database has more than one Entity View, do not show new Fields on these view unless a field created right from this view
P.S. Thank you for your patience and support 
.
