Quotas & training status
The Knowledge base page surfaces two systems that show you what’s happening right now: per-source training status and per-organization quota usage.
In this guide:
- Read training-status pills
- Read the quota bar
- Resolve a stuck source
- Decide what to do when you hit a quota
Training-status pills
Every knowledge source has a status that updates in real time:
| Pill | Meaning |
|---|---|
pending | Source queued; waiting for a training slot. |
crawling | Fetching pages (website sources). |
processing | Extracting text from a document or transcript. |
training | Building the embeddings index. |
trained | Live and answering. |
failed | Something went wrong; hover the pill for the error. |
re-training | Auto-retrain is mid-run; index will update when complete. |
The dashboard polls automatically — you don’t need to refresh. WebSocket events also fire on completion (see Auto-retraining for the toast notification).
Quota bar
At the top of the Knowledge base tab, a bar shows total quota used vs. allowance.
Screenshot: The knowledge-quota usage bar with current and allowance values.
Three colors:
- Green — comfortable headroom.
- Amber — past 75% used; consider what to prune or upgrade.
- Red — at or over the cap; new sources will refuse training until you free up space.
What counts toward quota
- Total trained content size (text bytes after extraction).
- Re-trained sources count once at their current size; previous versions are released when the new version completes.
- Unsuccessful (
failed) sources don’t count.
What does not count:
- Knowledge sources you’ve added but deleted.
- Failed crawls/uploads.
- Bot conversation messages — that’s a separate message quota tracked on the Usage analytics page.
Resolve a stuck source
If a source is stuck on pending for over 30 minutes:
- Refresh the page.
- If still pending, the queue may be full — wait. (Plans with high-priority training queues complete sooner.)
- If still pending after an hour, contact support.
If a source is failed:
- Hover the pill to read the error.
- Common errors: “0 pages crawled” (try Documents), “no transcript” (YouTube without captions), “timeout” (very large file — split it).
- Delete and re-add with the fix applied.
When you hit a quota
Three options:
- Prune — delete sources you no longer need. Click any source row and use the trash icon.
- Buy a knowledge expansion add-on — one-time or monthly capacity boost. → Add-ons
- Upgrade your plan — the cleanest path if you’re growing. → Plans, usage & pricing
Tips for quota efficiency
- Use websites over documents when possible. Websites de-duplicate naturally; uploading 50 PDF copies of the same handbook wastes quota.
- Split a giant document into focused snippets for the parts that actually answer questions. The full document is rarely necessary.
- Turn off auto-retrain for sources that rarely change. Each scheduled re-train consumes quota even when content is identical (websites can’t hash without re-crawling).
What’s next
You’ve finished the Knowledge base section. Next, learn the Playground.