FIN is 'stuck'/'clogged'/frozen - eternally loading uploaded content & can't ingest new content

Hi,

It sounds like your Fin sync is stuck in a "clogged" state because the high volume of small snippets has overloaded the ingestion queue, causing it to default to your cached 2025 data. To fix this, you’ll need to pause the sync and delete the problematic snippets to clear the "pipe," then force a refresh by uploading one small test file. For a long-term solution, stop using manual .txt snippets for database metadata; instead, use Data Connectors to link Fin directly to your JSON source via API. This allows Fin to "ping" your database in real-time, handling complex terminology and pricing updates automatically without the risk of freezing the system again.

Hi Nico, 

Thanks for your reply and suggestion - I’ll give it a try. However, one thing that I forgot to mention is that I organized all those ‘small’ text file snippets in a folder in Intercom. Its a folder in the knowledge section, containing ~125 snippets. If I try to open the folder containing the snippets, it loads for a very long time, after which it errors and pretends like the folder is empty - it displays zero of the uploaded files. Maybe I’m still able to pause the sync, but I’m not sure if I’ll manage to delete files that I’m not able to view. 😅

Any other ideas/suggestions are very very welcome 🙏

If you are unable to view the files in the folder due to the timeout error, you can regain control and delete them by using the Knowledge > Content tab instead of the specific folder view. Search for your snippets globally or use filters to isolate them in the main list, where you can then perform a bulk selection to delete them directly without needing the folder to load. If the UI remains unresponsive, you can bypass it entirely by using the Intercom REST API to fetch a list of all article IDs via a GET request to https://api.intercom.io/articles and then programmatically batch-deleting those specific IDs using DELETE requests.

To further ensure the UI loads correctly, perform a hard refresh (Command-Shift-R on macOS or Ctrl-F5 on Windows) or use an incognito window to clear any cached metadata or extension interference that might be triggering the "zero files" glitch.

If I use some AI to investigate, I get the following answer: 

If You Truly Need the Full 1GB (our metadata file is ~850MB), then Data Connectors alone won't work, and you'd want a hybrid approach:Data Connectors for the curated summaries (covers 90% of questions)Custom Action that calls your own search API for deep/specific lookups when Fin needs more detail (e.g., "what's the exact schema of table X including all 200 columns")This way, most questions get answered instantly from the synced content, and only edge cases require a live API call.Bottom line: Don't try to push 1GB into the connector. Curate it down to what's actually useful for answering user questions — you'll get better answers and stay within platform limits.

Thoughts on this? 

Yes, I agree to this approach. This hybrid model is the best way to handle large technical datasets because it keeps Fin fast while ensuring the answers stay accurate. By curating your core info into a "Knowledge Layer," you handle the 90% of common questions instantly without clogging the system with massive files. For the deep technical stuff, you can use Data Connectors to "ping" your own servers in real-time; this way, Fin is never "stuck in the past" because it's pulling live data on demand. To get things moving again, just clear out those old snippets using the Knowledge > Content tab, then set up a simple API endpoint for your metadata so Fin can fetch exactly what it needs without any lag.

Understood, thank you ​@Nico Magbiray 

We have removed all content that was ‘stuck’ / eternally spinning when it comes to syncing with FIN, as per your suggestion. We have paused all syncing + removed the content.

Now, ~24hrs later we are still unable to upload a small new ‘test’ file such that FIN is able to process it. It appears to still be stuck - any other ideas what could unclog it, or how we can somehow ‘reset’ the system?

Hi,

If your Fin AI Agent remains stuck even after deleting the old content, it’s likely that background tasks are still trying to process the original "ghost" data. Follow these steps to clear the ingestion queue and force a fresh sync:

1. Force a System Reset

• Fully Disable Fin: Go to Fin AI Agent > Deploy and turn it completely off for all channels (Chat, Email, etc.).

• Wait for the "Quiet Period": Leave Fin disabled for 15–30 minutes. This allows the backend queue to clear out any pending tasks associated with the deleted files.

• Upload a "Trigger" File: Before re-enabling, upload one very small, fresh text file (e.g., "Reset Test 2026").

• Re-enable and Test: Turn Fin back on and use the Preview User tool in the Test Fin tab to see if it pulls information from your new test file.

2. Clear the Cache

• Start a New Conversation: Fin often caches answers within a specific conversation thread. Always start a brand new test chat to ensure you aren't seeing old, cached data from a previous "stuck" session.

• Check Sync History: Go to Fin AI Agent > Train > Content > View sync history. If you see a status like "Failed" or "Processing" on your old 1GB upload, it is still blocking the pipe.

If It Is Still Not Syncing

If your status still doesn't show "Synced" within an hour, Intercom Support will need to manually flush your ingestion queue. To help them, you’ll need to provide a Conversation ID where the issue is occurring:

• Find via URL: Open a conversation in your Inbox and look at the end of your browser's URL. The number after /conversations/ is your unique ID (e.g., .../conversations/215472322478539).

• Find via Sidebar: In the conversation view, open the Attributes app in the right-side panel; the Conversation IDis listed there.

Let me know if that worked.

We have the same issue; I updated the help center, but Fin is still using the old data. This is misleading for the customer because Fin is essentially lying to them in the chat when they're searching for something specific.