Knowledge bases
A knowledge base gathers your business documents (manuals, policies, FAQs, product sheets) so your AI assistants and the AI steps of your flows answer with them. Search works by meaning, not only by exact words: if a customer asks “how long until I get my money back?”, it finds the “refund period” section.
What you gain
Section titled “What you gain”- Answers you can back up. Every fact in the answer carries its citation: which document and which page it comes from. You check it in one click, and the customer knows where it comes from.
- Fewer made-up answers. If the answer is not in the documents, the assistant says so. And if an assistant answered without citing anything, you see it flagged as Sin fuente (No source) so you can review it.
- One place to maintain. You update the document in the base and every assistant and flow in the environment answers with the new version.
Where it is
Section titled “Where it is”In the menu, Centro › IA › Conocimiento (Center › AI › Knowledge). Bases belong to the environment you choose at the top right. To see them you need permission to view assistants; to create, upload or delete, permission to edit them.
Create a base and upload documents
Section titled “Create a base and upload documents”- Click Nueva base (New base).
- Enter the Nombre (Name), for example
Product documentation, and a Descripción (Description) of what the documents are about: the assistant reads it to know when to search there. - Click Crear base (Create base) and open the base.
- Drag the documents onto the upload area, or pick them: PDF, Word (.docx), Excel (.xlsx), text, Markdown, CSV or JSON, up to 50 MB each.
Each document goes through the antivirus and moves through En cola (Queued), Procesando (Processing) and then Listo (Ready) for the assistants, within seconds. If something fails, it stays in Error with the reason, for example:
- “No text was found. If it is a scanned PDF (image), run it through OCR first.”
- “Almost no page has readable text: it looks like a scanned document.”
- The document is larger than the plan allows (see Plan limits).
Volver a procesar (Reprocess) tries again; if it fails again because of a limit, the assistant keeps answering with the previous version. Quitar documento (Remove document) takes it out of the base. A file identical to one already in the base is not uploaded twice.
Test the search
Section titled “Test the search”In Probar búsqueda (Test search), type a question the way a customer would and you will see the passages the assistant would receive. It is the fastest way to know whether a document answers what you expect before putting it in front of customers.
Use the base
Section titled “Use the base”- In an assistant: in its Configuración (Settings), Conocimiento (Knowledge) tab, Base de conocimiento (Knowledge base) card, check the bases it consults (saved as you check them). Before answering about those topics, the assistant searches them with Buscar en documentos (Search documents) and marks each fact with its source; if the answer is not there, it says so.
- In a flow: in an AI step, Conocimiento y modelo (Knowledge and model) section, choose the Bases de conocimiento (Knowledge bases) and Qué buscar (What to search). If a base no longer exists in the environment, or there is nothing to search, the step becomes an incident with the reason.
Citations
Section titled “Citations”Every fact that comes from your documents is marked with a number in brackets, such as [1] or [2, 4], and its sources appear below the answer: “[2] Manual.pdf · p. 4”. Hovering shows the cited passage, and anyone with permission opens the document right at that page.
Zoan Cloud builds the citations from what it actually found in your documents, not from what the model writes: a number that does not match a delivered passage is removed, and a file name written by the model does not count as a source.
You see them in the assistant test, in Conversaciones (Conversations), in the assistants’ flow mode and in the log of each case of a flow (Fuentes, Sources, section).
No source
Section titled “No source”In assistants with knowledge bases, an answer without citations carries the Sin fuente (No source) flag, with the reason on hover:
- “It searched the documents but the answer cites no passage: review it.” The assistant searched, but what it answered does not come from what it found.
- “It did not search the documents for this answer.” It answered without consulting the bases.
Review those answers in Conversaciones: they usually point to a missing document or an assistant instruction worth adjusting. Greetings and the fixed questions of flow mode are never flagged.
What a web chat visitor sees
Section titled “What a web chat visitor sees”In the assistant settings, Fuentes en el chat web (Sources in the web chat) decides what someone chatting with your assistant on your website sees:
| Option | What the visitor sees |
|---|---|
| Nombre del documento y página (Document name and page, default) | Which document and page the answer comes from, without being able to open it. |
| Nombre, página y enlace para abrir el documento (Name, page and link to open the document) | Also a link to open the cited document at that page. It only opens the documents cited in their conversation. Use it only with bases of public documents. |
| No mostrar fuentes (Don’t show sources) | Only the answer. You still see the sources in Conversaciones and in the test. |
The visitor never sees internal passages or identifiers. On WhatsApp, the numbers are removed from the text and the answer ends with a line “Fuente: document (p. N)”. Changing this option requires permission to publish the assistant, and it is recorded in the audit log.
Plan limits
Section titled “Plan limits”| Free | Starter | Pro | |
|---|---|---|---|
| Bases per environment | 2 | 10 | 50 |
| Documents per base | 25 | 200 | 1,000 |
| Indexed passages (whole organization) | 5,000 | 50,000 | 500,000 |
| Passages per document | 1,000 | 3,000 | 5,000 |
| Documents uploaded or reprocessed per hour | 30 | 120 | 600 |
A passage is a chunk of about 1,500 characters, roughly half a page. Enterprise plans have at least the Pro caps. The Conocimiento page and Gobierno › Consumo de IA (Governance › AI usage, Bases de conocimiento card) show how much of each limit you have used; past 80 % they are flagged as a warning.
When you reach a limit, Zoan Cloud tells you which one and what to do, for example: “The plan allows 25 documents per base. Delete some, use another base or change plans.” A document that produces more passages than allowed is not truncated: it is rejected with a notice, so you can split it into parts. If you downgrade your plan, nothing is deleted; you just cannot add more until you are within the limit.
Fair use
Section titled “Fair use”To keep document processing even for everyone:
- Uploads per hour. Uploads and reprocessing count toward the plan’s hourly cap. When you reach it, the message says in how many minutes you can continue. A rejected upload does not count.
- Processing in turns. Each organization’s documents are processed one at a time and in fair order across organizations: if you upload many at once, some stay En cola (Queued) for a while. You don’t need to do anything.
- Search tests. Probar búsqueda allows 30 queries per minute per person.
Gobierno › Auditoría (Governance › Audit) records who created, edited or deleted a base, who uploaded, reprocessed or removed a document, when a plan limit blocked something and who changed an assistant’s bases or web chat sources. Each query from an assistant or an AI step to the bases is recorded as Consulta a las bases de conocimiento (Knowledge base query), with the bases, how many results there were and from which documents and pages, never with the text of the passages. The question text is only stored, trimmed and with personal data replaced, for assistants that mask personal data; for the rest, only its length.