Knowledge sources
Training status
What each source status means, how to fix sources that fail, and how retraining, deleting and storage limits work.
Before your agent can answer from a source, OrbitBase has to read it and train on it. Every source shows a status so you always know what your agent can use right now, and what needs your attention.
Statuses
| Status | What it means | Does your agent use it? |
|---|---|---|
| Uploaded | The document is saved but hasn't been trained yet. Shown as Not trained in the guided setup. | No |
| Training | OrbitBase is reading and training on it. This usually takes a few minutes. | No |
| Trained | Ready. Your agent answers from it. | Yes |
| Error | Training didn't finish. The reason is shown on the source. Shown as Failed in the guided setup. | No |
Your agent only answers from sources marked Trained. While a source is retraining, or if its last training failed, your agent doesn't use it until it's back to Trained.
The Sources overview card sums this up under Training status: how many sources are Trained, In progress and Needs attention.
When training starts
- Websites, free text and Notion start training as soon as you add them or save a change.
- Documents uploaded on the Documents page wait as Uploaded until you click Train or Train all. Documents uploaded during the guided setup start training straight away.
Train and retrain
You can start or repeat training whenever you need to:
- Train — in a source's row menu, or in the Actions menu when you open it. Trains that one source.
- Train all — at the top of the list. Trains every source on that page that's waiting or failed.
- Refresh from the site — for website pages. Re-reads the page from your site and retrains it if it changed.
- Sync again — for Notion. Brings in new and changed pages.
Retraining replaces what your agent learned from that source with the latest version, so repeating it never creates duplicates.
Automatic refreshes
- Website pages are re-read every 7 days.
- Notion pages are re-synced every day.
Only pages that have changed are retrained.
Fix sources that failed
The reason training failed is shown under the source's name in the list, and at the top of the source when you open it. Common reasons:
| Message | What to do |
|---|---|
| We couldn't find any text in this file. | The file is probably a scan. Export it with text recognition (OCR) turned on and upload it again. |
| This document has no readable text to train on. | Upload a version with selectable text. |
| No readable text found on that page. | The page's content may load with scripts, or be mostly images. Add the content as free text instead. |
| That site's robots.txt asks crawlers not to read this page. | Allow the page in your site's robots.txt, or import a page it allows. See Websites. |
| This page has no text to train on. | A Notion page with no written content. Add text to it in Notion, then sync again. |
| Not trained — this page would take your plan past its training storage. | Upgrade, or remove a source you no longer need, then train it again. |
| Training stopped unexpectedly. Try again. | Training was interrupted. Click Train to try again. |
| Couldn't start training. Try again in a moment. | A temporary problem. Wait a moment and click Train again. |
If a source stays in Training for much longer than usual, OrbitBase eventually marks it Error with "Training stopped unexpectedly. Try again." so you can retrain it.
After fixing a source, ask a question it covers in the Playground to confirm your agent picks it up.
Delete sources
Owners and admins can delete sources. Open a source's row menu and click Delete, or select several and click Delete — the button shows how many you've selected. Deleting removes the source and everything your agent learned from it, straight away. This can't be undone.
- Website pages you delete won't be re-added from your sitemap. Add a page by its link to bring it back.
- Notion pages you delete won't be imported again by future syncs. Disconnect and reconnect Notion to bring one back.
Storage limits
Each plan includes a set amount of knowledge storage, shared by every workspace in your organisation. It's measured on the text OrbitBase keeps from each source, not on file sizes.
- Before you add anything, OrbitBase checks there's room. If there isn't, it tells you how much storage is left and doesn't add the source.
- During an automatic refresh or sync, a page that would take you past your limit isn't trained, and is marked with an error explaining why.
- To make room, delete sources you no longer need, or upgrade your plan in Settings → Billing.
Watch your usage on the Sources overview card under Training data, or in Settings → Usage. See Plans and limits.