Brand knowledge base

Kreel keeps a knowledge base for each brand: a typed profile, a few narrative sections, a list of named competitors, and a running list of learnings. A connected AI agent reads this before it analyzes your data or publishes a report, so its answers reflect who your brand actually is, not just the numbers in front of it. You can fill it in by hand in the dashboard, or let your connected agent interview you and write it for you.

No Kreel-run AI

Kreel does not run its own AI to write or infer this content. It either comes from your own connected agent, using your own tokens, or from what you type by hand in the dashboard.

Where to find it

1. Sign in to Kreel at app.kreel.ai.
2. In the left navigation, click Brand.

The page is titled “Brand Knowledge Base.” If nothing has been filled in yet, you see an empty state that points you at your agent: connect one and run the onboarding prompt, or fill in each tab by hand yourself. Six tabs, in order: Profile, Positioning, Ideal customer, Brand voice, Competitive landscape, Learnings.

What’s in it

Four kinds of content, spread across those six tabs:

• Profile - a small set of typed fields: vertical/category, price tier, margin band, and optional target ROAS, CPA, and AOV.
• Sections - free-form markdown narrative, one each for Positioning, Ideal customer, and Brand voice.
• Competitors - a structured list of named rivals, each with an optional handle/URL and a one-line differentiator.
• Learnings - durable insights your brand accumulates over time, each with a topic, content, and optional evidence.

Profile and Sections never expire or get capped. Only Learnings have a cap, so a handful of one-off insights can never crowd out your positioning doc.

Profile

The Profile tab holds:

• Vertical / category - free text, for example “skincare.”
• Price tier - a dropdown: value, mid, premium, or luxury.
• Margin band - a dropdown: lean, moderate, or healthy.
• Target ROAS, Target CPA, Target AOV - optional numeric targets.

Click Save brand profile to save.

Positioning, Ideal customer, Brand voice

Each of these tabs holds one markdown section. Click Edit (or Add if the section is still empty) to open a text area, write or paste markdown, then Save. Each field starts with placeholder text showing the kind of content it expects.

Competitive landscape

This tab combines a Competitive landscape narrative section with a table of named competitors: Name, Handle / URL, Differentiator. Click Add competitor to add a row, or use Edit / Remove on an existing one. Removing a competitor is immediate - there is no confirmation dialog. If you (or your agent) add a competitor whose name is already there, only differing by case or spacing, Kreel treats it as the same competitor and updates that row instead of creating a duplicate.

Learnings

The Learnings tab shows a badge like 12/50, your active learning count against the soft cap of 50. The badge switches to a “danger” style once you reach 80% of the cap (40/50), with a hint to archive stale learnings. The cap is a signal, not a limit - Kreel never blocks you or your agent from recording a 51st learning.

Click Add learning and fill in Topic, Content, and optional Evidence. If you save a learning whose topic already has other active learnings, Kreel still saves it, then shows a banner naming those entries so you can edit one of them instead of keeping a near-duplicate. Each learning row shows its content, its evidence (if any), how long ago it was recorded, and Edit / Archive buttons.

Archiving learnings

Archiving a learning is a soft delete: it drops out of your active count and out of what your agent reads by default, but the record itself is kept, not deleted. There is currently no way to bring an archived learning back through the dashboard or through your agent. If you archive one by mistake, record it again as a new learning.

Let your agent read and write it

Once you connect an agent over MCP, it can read the whole knowledge base in a single call and write to it through six dedicated tools. See Brand context tools for the full parameter list. Each write tool asks for your approval before it runs, the same as any other change your agent makes in Kreel.

The dashboard’s empty state also points you at a guided interview: connect your agent and ask it to run the onboarding prompt for this brand. It picks the brand (if you have more than one), reads what’s already in the knowledge base, gathers evidence from your connected performance data, asks you only about the gaps it can’t infer, and writes nothing without your confirmation. It’s safe to run more than once - each pass just fills in whatever is still missing. See the prompt library for what it does step by step.

An empty knowledge base still works

An agent that reads an empty or partial knowledge base still answers your questions, it just has less to anchor on. Filling it in, by hand or through the interview, makes every analysis and report more specific to your brand.

Not part of the query tool

Profile, sections, competitors, and learnings are qualitative, not tabular, so they aren’t available through the query tool. Your agent reads them as a whole through the brand context tool instead.

If you use the CLI, kreel context shows the same knowledge base, read-only. See CLI.

Need help?

Email support@kreel.ai.