Knowledge Base

A knowledge base is a collection of your documents (files or crawled web pages) that CloudThinker searches when it answers. Add your reference docs once, and answers in chat and incident investigations draw on your own procedures instead of generic advice.

How it works

You create a knowledge base and add files or a website URL.
CloudThinker indexes each document in the background. Each one shows a status: Pending, Processing, Completed, or Failed.
CloudThinker decides when to search from each knowledge base’s title and description, then searches the completed documents by meaning, not exact keywords, and answers from the passages it finds. Resolve can also use them as evidence during incident triage.

Every agent, built-in or custom, can search every knowledge base in the workspace, and searches all of them by default. Name a knowledge base in your request to limit the search to that one. Naming one in a custom agent’s instructions steers which it searches; it doesn’t block the others. Knowledge bases belong to one workspace. Everyone in the workspace can read and use them; agents in other workspaces can’t. Use a knowledge base for reference material to look up, and a skill for procedures agents follow every time. See when to use a skill, or Teach CloudThinker for every surface that holds your team’s context.

Prerequisites

The workspace Admin role to create knowledge bases and add or delete documents. Developers can browse them and ask questions. Viewers can browse them but can’t chat. See roles.

If Knowledge is missing from More → Workspace settings, the Knowledge Base module is turned off. Ask CloudThinker support to turn it on, or, if your company runs CloudThinker itself (self-hosted), ask your platform team. Turning it off hides the page and stops new ingestion and search, but keeps stored documents.

Create a knowledge base

Open Knowledge

Go to More → Workspace settings → Knowledge.

Click Create KB

Create KB sits to the right of the search and filter bar, above the list. The Create Knowledge Base dialog opens.

Fill in the details and click Create

Field	What to enter
Title	A name people will use in prompts, such as `Payments runbooks`. Up to 255 characters.
Description	What’s inside and when to use it, such as “On-call runbooks for the payments API, queue workers, and Postgres.” CloudThinker reads this to pick the right collection. Up to 1,000 characters.
Tags	Optional labels to filter the list, such as `oncall` or `payments`.

Success state: the new knowledge base appears in the list with 0 documents.

To change the title or description later, open the row’s menu in the list and click Settings. Keep the description current: CloudThinker uses it to decide when to search.

Add documents

Open the knowledge base, click Add Documents, and choose a tab.

Files
Website

Drag files onto the drop area, or click it to select them.
Click Upload N files.

Limit	Value
Formats	PDF, DOC, DOCX, TXT, MD, CSV, XLSX, XLS, ODS. Other formats, such as PPTX, HTML, and images, aren’t accepted.
Size per file	10 MB
Files per upload	10

CSV and spreadsheet files are converted to text tables, then searched by meaning like any other document. Uploads also count toward your workspace’s storage quota.

Enter the Website URL, for example https://docs.example.com/runbooks.
Turn on Deep Crawl to also import pages it links to on the same domain. Leave it off to import only that page.
Click Add Website.

Pages load in a headless Chromium browser, so content that JavaScript renders is read. A deep crawl stops at 20 pages and two links deep, and you can’t change that in the app. For a larger site, add several start URLs, one per section. The URL must be public http or https. Internal addresses and URLs with a username or password in them are rejected.An import is a one-time copy, and there’s no scheduled refresh. A deep crawl adds one document row per page it imports. To pick up changes, delete the site’s rows and add the URL again.For internal docs behind a login, export the pages to PDF or Markdown and upload them on the Files tab. For Confluence, an Atlassian connection lets agents search pages live instead.

Success state: every document in the list shows Completed. Only completed documents are searchable. Large files and deep crawls take longer.

Ask questions from your documents

Ask in plain language. Ask about the content you need, not the document’s title, because CloudThinker searches by meaning. Name the knowledge base when you want to search only that one:

What's the escalation path for a Sev 1 on the payments API?
Using the Payments runbooks knowledge base, how do I fail over the primary Postgres?
Summarize the backup and restore procedure for our Redis clusters

Success state: the agent’s activity shows a Knowledge Base (document search) step, and the answer follows your document’s steps and wording. Answers don’t carry inline citations. To check what the answer drew on, expand the Knowledge Base step in the activity, or ask “Which document did you use?”.

Keep it accurate

Replace, don’t stack. When a runbook changes, delete the old document and upload the new one. CloudThinker can’t tell which of two copies is current.
Use clear headings and short sections. Search returns passages, so a focused section makes a better answer than a long wall of text.
Start small. Load the ten documents your team opens most during incidents, then add more as questions come up.

To delete a document, use Delete in its row. Its search index entries are removed, so it stops being searchable right away, and its size comes off your storage quota. To delete a whole knowledge base, use Delete from its row in the list.

Deleting a knowledge base permanently deletes every document in it. This can’t be undone.

Data and access

Everyone in the workspace can read every knowledge base. Don’t upload documents only some members should see; put them in a separate workspace.
Never upload secrets. Keep API tokens and passwords in Credentials, which agents receive only when they run commands.
Passages reach the model. The passages a search returns become part of the conversation, so they go to CloudThinker’s LLM provider, or to your own with BYOK. With BYOK, a call your key can’t serve falls back to CloudThinker’s models, so passages can still reach CloudThinker’s provider. See When a key stops working.
For a security review, see Security and Trust for what CloudThinker stores and how it’s protected.

Troubleshooting

CloudThinker doesn't use my documents

Check the document status. Only Completed documents are searchable.
Ask a specific question about the content, and name the knowledge base to narrow the search.
Make sure you’re chatting in the workspace that owns the knowledge base.
Give the knowledge base a description that says what’s inside.

A document shows Failed

Delete it and upload it again. If it fails again, export it to a simpler format (for example PDF or TXT) and retry.

Files won't upload

Each file must be 10 MB or smaller, with at most 10 files per upload.
The format must be one of PDF, DOC, DOCX, TXT, MD, CSV, XLSX, XLS, ODS.
“Storage quota exceeded. Cannot ingest more documents.” means the workspace is out of knowledge base storage. Delete documents you no longer need. The quota is set by your plan and isn’t shown on the Usage page, so to raise it, contact CloudThinker support.

A website import returns nothing

Open the URL in a private browser window. If it needs a login or VPN, CloudThinker can’t reach it.
Some sites block crawlers. Try a different starting page.
Turn off Deep Crawl and import one page first to confirm the site can be read.

I can't see Create KB or Add Documents

You need the workspace Admin role. Ask a workspace admin to add the documents or change your role.

Connections

Customize

Approvals

Workspace settings

Admin Settings

Artifacts and sharing

How it works

Prerequisites

Create a knowledge base

Add documents

Ask questions from your documents

Keep it accurate

Data and access

Troubleshooting

Next steps

Custom skills

Runbooks

​How it works

​Prerequisites

​Create a knowledge base

​Add documents

​Ask questions from your documents

​Keep it accurate

​Data and access

​Troubleshooting

​Next steps

Custom skills

Runbooks

How it works

Prerequisites

Create a knowledge base

Add documents

Ask questions from your documents

Keep it accurate

Data and access

Troubleshooting

Next steps