Button to AI Analyze a set of entities (child entities, or data view)

Ideas & Features
1

See the following feature in Google Drive, which uses Gemini,
that automatically gives a summary with analysis of a folder ant its containing files!

image
image1700Ă—965 130 KB

We want that in Fibery also, right?
That is, an AI button to do that for a relationship collection list/table view.

2

Have you tried making a button to do exactly this?

Or using Build AI?

3

It worked eventually using a button, but this definitely needs polishing in Fibery, plus some faster RAG implementation, as well as showing the thinking process.

Firstly i tried using the button “Ask AI to generate markdown template”, which resulted in scripts using the [ai] tag instead of an [ai-agent] prompt.

image
image1077Ă—502 35.4 KB

Case Study

I successfully automated a risk analysis for a Dossier with 20 linked Documents, but the native “Generate Prompt” button led me down the wrong path.

1. The Trap: Outdated Defaults
The “Ask AI to generate template” button creates legacy scripts using the [ai] tag and text loops.

  • Result: “Context Stuffing.” It forces the AI to read everything linearly, hitting token limits and failing with large datasets.

2. The Fix: [ai-agent]

I manually switched to the [ai-agent] tag with natural language instructions (“Read all linked documents…”).

  • Result: The Agent autonomously retrieved and processed the data, bypassing token limits successfully.

3. The UX Gap: “Black Box”

  • Latency: The Agent took 5 minutes to “reason” and loop through the data.
  • No Feedback: Automations lack a “Thinking…” state. The UI freezes, looking like a crash.
  • Workaround: I had to add a manual step to update a text field to “:brain: AI is thinking…” just to inform the user.

Feature Requests

  • Native Vector Actions (Fast RAG): Agents are overkill for simple summaries (5 mins) and the old loop method breaks at scale. We need a direct Query Vector Index action for fast, deterministic retrieval (seconds vs. minutes).
  • Native UI Feedback: Add a standard “Processing” indicator or streaming text for long-running AI automations so we don’t need manual workarounds to prevent the “frozen screen” experience.
  • Smarter Defaults & Guidance: Don’t simply replace the old loop with the slow [ai-agent]. Instead, the builder should detect large collections and suggest the appropriate tool (Vector Search vs. Agent) to avoid performance pitfalls.
1 Like
4

How large were the 20 documents you were trying to analyse?

Which model did you set in the prompt?