> ## Documentation Index
> Fetch the complete documentation index at: https://docs.keloa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Managing sources

> Recrawl, retry, reassign, delete — keep your knowledge clean.

The **Knowledge** list is where you manage every source. Filter, search, and take action from one page.

## List view

Each row shows:

* **Type** (website / file / snippet / Q\&A)
* **Name**
* **Attached agent**
* **Status** (queued / syncing / synced / failed)
* **Last synced**

Filter by type or status at the top. Search by name.

## Recrawl a website source

Open the source → **Recrawl**. Keloa refetches every page and re-indexes. Old chunks are swapped for new atomically — there's no "gap" where the agent loses the knowledge.

Use this when you've updated pages, added new ones, or a customer gets a stale answer.

## Retry a failed source

If the status is **failed**, open it to see why. Common causes:

* Domain unreachable during the crawl → **Retry**
* File parsing error (corrupted PDF) → re-upload
* Cap exceeded → remove unused sources or upgrade plan

Retry queues the whole source again.

## Retry a single page

Inside a website source, each page has its own status. If one page failed (e.g. a 5xx at the time), click **Retry** on just that row without rebuilding the whole crawl.

## Reassign to another agent

Open the source → **Agent** dropdown → pick a different agent → save. Re-indexes against the new agent in seconds. Useful when splitting one agent into two.

## Delete a source

**⋯ → Delete** on the row. Immediate:

* Chunks are removed from the vector index.
* Raw file (if any) is scheduled for deletion.
* The agent will no longer see this content on the next message.

## Bulk delete

Select rows with the checkboxes → **Delete selected**. Great when cleaning up after a reorg.

## Storage budget

The bar at the top of the page shows characters used vs. your plan cap. When you hit the cap, new sources fail with "over cap" — delete unused sources or upgrade.

Vector count (under the bar) is a rough indicator of index size; it drives search latency, not your bill.

## Troubleshooting

| Issue                                   | Fix                                                                  |
| --------------------------------------- | -------------------------------------------------------------------- |
| Source stuck at **queued**              | Worker isn't running. Wait 1–2 minutes, or contact support.          |
| Source stuck at **syncing** for hours   | **Retry**. If it fails again, check the URL or file.                 |
| Answer doesn't reflect a page I updated | **Recrawl** the source.                                              |
| Too many pages crawled on a small site  | Use a narrower starting URL (e.g. `/products` instead of `/`).       |
| Sensitive info leaking into replies     | Delete the source containing it, then **Recrawl** remaining sources. |