Knowledge base

Train an agent on your documents and website so it answers only from your content, never made up.

An agent's knowledge base is the content it answers from. Add your documents and website pages, and the agent grounds every reply in that material instead of guessing. Pair it with the Answer only from my knowledge guardrail and the agent declines, or hands off, anything your content doesn't cover, rather than inventing an answer. You manage it all from the Knowledge Base tab of the agent editor.

Adding knowledge

Two ways to add a source, both from the Knowledge Base tab:

Upload a file: a document or an image.
Import from a URL: enter a website address and AnyBot crawls it, adding the pages it reads to the knowledge base. Connect the Website Crawler integration to crawl a whole site and optionally keep it fresh on a schedule (daily, weekly, or monthly).

Source	What it ingests
Documents	PDF, Word, Excel, CSV, JSON, HTML, Markdown, plain text
Images	JPEG, PNG, GIF, WebP (read and described by vision AI, then indexed as text)
Website	Crawled by URL, extracted, broken into sections, and added to the knowledge base

The one-time site scan the sign-up wizard runs to build your business profile is a separate step from the agent's knowledge base.

The ingestion pipeline

Adding a source doesn't make it answerable instantly. Each one is processed in the background before the agent can use it:

   Upload a file / import a URL
              │
              ▼
        Pending ──▶ Processing ──▶ Ready
                        │        (chunked, embedded, searchable)
                        ▼
                     Failed ──▶ delete + add again

Pending: queued for ingestion.
Processing: the source is split into chunks and embedded so the agent can search it by meaning; images are read and described by vision AI first.
Ready: searchable. The agent can now cite it.
Failed: something went wrong (an unreachable URL, an unreadable file). Delete the source and add it again; there is no separate retry button.

The Knowledge Base tab shows each source's live Status and a Chunks count, so you can watch a large document move through the pipeline.

Shared library vs. per-agent

Every source has a scope that decides which agents can reach it:

Scope	Who can use it
Space library	Shared, available to every agent in the space
Agent	Private, only the agent you added it to

Upload a policy PDF or price list once at the space library level instead of re-uploading it to every agent. The Knowledge Base tab tags each row with its scope, so you can tell at a glance whether a source is the agent's own or inherited from the shared library. Manage the shared library from the space-level Knowledge page.

Grounded answers

Knowledge is only half the story. The Answer only from my knowledge guardrail is what keeps the agent honest. With it on, the agent answers strictly from what you added and politely declines off-topic questions; a low-confidence answer can hand off to a human instead. Rehearse this in the live test chat before going live: ask something you know is covered, then something you know isn't, and confirm the agent cites the first and declines the second. Tune the behavior on the Behavior & guardrails page.

Troubleshooting

A source is stuck on Failed. Delete it and add it again. For a URL, check the page loads publicly (no login wall) and isn't a redirect.
The crawl missed some pages. Pages behind a login or that block crawlers are skipped. Confirm the pages load publicly; connect the Website Crawler integration for scheduled re-crawls.
The agent ignores a source. Confirm its status is Ready, not Pending or Processing, and that it's scoped to (or shared with) that agent. If the agent still won't answer from it, check the answer-only guardrail and phrasing in the test chat.

FAQ

What file types can I add? PDF, Word, Excel, CSV, JSON, HTML, Markdown, and plain text, plus images (JPEG, PNG, GIF, WebP) that are read by vision AI. Your operator sets the size cap and the allowed list.

Can it read my whole website? Yes, enter your site's URL and AnyBot crawls it into the knowledge base. Connect the Website Crawler integration to keep it refreshed on a schedule.

Will it make things up if it doesn't know? No. With the answer-only guardrail on, it answers only from your content and declines or escalates when it can't.

Do all my agents share one knowledge base? Only what you scope to the space library. Anything added at the agent level stays private to that agent.