3. Build a knowledge base

Turn the uploaded documents into a curated, versioned knowledge base for your agent

Raw uploads are not the best input for an agent: they repeat each other, mix formats and bury the facts that matter. A knowledge base (Context) is a curated file tree an AI builder makes from your sources. Each publish creates a version, and agents read that published snapshot instead of the raw files.

In this part you build a knowledge base from the folder you uploaded in part 2 and publish it.

Before you start

• You completed 2. Upload files.

A knowledge base is built from files: workspace files, personal files, agent outputs, an agent’s own files, or an external source. Published analytic models are not a knowledge-base source. You connect the model from part 1 to the agent directly in part 4, and the agent uses both side by side.

1. Create the knowledge base

1. Open Context

   Click Files on the left navigation rail. In the Files & context panel beside the list, click All contexts under Context, then click New context.

2. Name it

   Field	Value
   Name	A name agents and people will recognize.
   Slug	Lowercase. Agents find the knowledge base’s files under this name. It cannot be changed later.
   Description	What the knowledge base covers.

3. Pick the sources

   Under Raw data, select the folder you uploaded in part 2. Selecting a folder keeps tracking it, so a later rebuild picks up files added since.

4. Create

   Click Create. The context builder opens.

2. Build it with the AI builder

The builder works in its own sandbox. The selected files are copied in read-only as inputs; nothing you build is saved until you publish.

1. Explore the sources

   Start with the suggested prompt What is in the sources? The builder summarizes the documents it found.

2. Describe the knowledge base

   Tell the builder how to organize the content for the agent.

   Example prompt

   Build a knowledge base an analyst agent can rely on:
   - one file per topic, with short headings
   - a glossary of business terms and how each metric is defined
   - the policies and rules that affect how numbers are interpreted
   - an index.md that says what is in each file
   Leave out duplicates and anything outdated.
   
3. Review the result

   Check the generated file tree and ask for changes until it reads well.

3. Publish

1. Click Publish context in the header. The button shows how many changes you are publishing.
2. In Publish this context, add Release notes (optional), for example First version from the reference documents.
3. Click Publish.

The knowledge base now has a live version. Open its detail page and go to the Versions tab: the live version carries the Live badge.

A knowledge base is a snapshot, not a live link. When the documents change, open the builder and publish again. To roll back, click Roll back to this version on an earlier version; it is published again as the newest version.

Optional: keep it up to date with an automation

On the knowledge base’s Automations tab, click New automation. Enter a Name, describe the job in What should it do?, add a Scheduled trigger, and click Save automation. Scheduled runs only update the draft; they never publish, so you review and publish the result yourself.

What you built

• A published knowledge base built from your documents, with version history.

You attach it to an agent in the next part. Learn more in Build a knowledge base and Use a knowledge base.

Previous: Upload files

Upload the reference documents.

Next: Build an agent

Put the model and the knowledge base to work.