Connecting Your Knowledge Base: The Grounding Layer
Your AI agents can sound like your brand, but without access to your proprietary data, they're still guessing at the details that actually matter. In this lesson, Rafaël shows you how to connect Knowledge Bases so every Workflow and agent you build is grounded in the information only your team owns.
What You'll Learn
Why your AI needs more than a brand identity
Your Brand Kit captures who you are, but your AI agents also need access to what you know. Most teams generate knowledge across dozens of tools: product docs, customer interviews, research, competitive intel, strategy decks. That knowledge is often scattered and invisible to the AI tools that need it most.
Knowledge Bases solve this by collecting your unique context and making it available to every Playbook you build inside AirOps.
What frontier knowledge is and why it matters
Frontier knowledge is the information only you and your brand possess. AI models can't produce it on their own because it doesn't exist in their training data. This is the information that differentiates your brand: original research, customer stories, product positioning, competitive analysis.
Content with original research and named expert bylines positively correlates with higher citation rates. When your Knowledge Bases contain frontier knowledge, every piece of content you produce is grounded in information your competitors cannot replicate.
Where your frontier knowledge lives
A Knowledge Base connects to all of it:
- Product documentation and help center articles
- Gong call transcripts where prospects describe their problems in their own words
- File uploads like strategy decks, playbooks, onboarding guides, and PDFs
- Your site map with published blogs and pages
- Competitor research
- Google Drive folders with positioning docs, customer success playbooks, and more
How semantic search works inside Knowledge Bases
Knowledge Bases use semantic search, which means retrieval based on meaning rather than exact keywords. When a Playbook asks for content about reducing customer churn, the Knowledge Base returns your most relevant material even if those exact words never appear in the source documents.
This is the same retrieval-augmented generation (RAG) pattern that powers AI search engines. The difference is that you're grounding your Playbooks in your own data instead of what's available on the open web.
The result: instead of giving an LLM vague instructions like "write in our style about our product," you give it your actual documentation, your best content, and your customer stories. Output goes from generic to grounded.
Setting up a site map Knowledge Base
Rafaël walks through creating a Knowledge Base from a sitemap:
- Navigate to Knowledge Bases and create a new one (he names it "Site Map")
- Connect a site map URL and specify URL filters (for example, URLs must contain "blog")
- You can also exclude URLs by specifying what they should not contain
- Select whether to scrape full content or metadata only
- Schedule recurring data syncs (for example, every Friday) so your Knowledge Bases stay refreshed automatically
Testing semantic search retrieval
Rafaël demos search against a help docs Knowledge Base with 765 items. Searching "how can I make my site more readable to AI agents" returns ranked, relevant chunks almost immediately, including a help center doc about using Markdown for AI agents.
Adding unpublished knowledge from Google Drive
Your sitemap captures what you've published, but your unpublished knowledge is even more valuable. Call transcripts are a gold mine because prospects describe their pain points in their own words. You can also bring in product positioning docs, competitive analysis, and customer success playbooks.
Rafaël creates a Knowledge Base called "call transcripts" and connects a Google Drive file by sharing access via email and importing the link. You can schedule syncs for Google Drive sources too, keeping Knowledge Bases current on a recurring basis.
Using metadata to unlock dynamic filtering
When you add documents to a Knowledge Base, you can attach metadata keys like:
- "type" (transcript, doc, playbook, call)
- "audience" (enterprise, SMB)
- "product" to specify product lines
These tags unlock dynamic filtering in Playbooks. For example, a blog Playbook targeting a specific buyer can pull only enterprise-tagged knowledge automatically.
How Knowledge Bases and Brand Kit work together
At this point, your Knowledge Bases contain knowledge unique to your brand, and every Playbook in AirOps can retrieve it exactly when needed:
- Your Brand Kit ensures output sounds like you
- Your Knowledge Bases ensure output is grounded in what you know
- Both update continuously. Brand Kit stays current through MCP, and Knowledge Bases stay current through scheduled syncs and new imports
Key takeaways
- Frontier knowledge is your competitive moatContent built on original research and named expert bylines correlates with higher citation rates in AI search. Your competitors can access the same LLMs you can, but they can't access your call transcripts, internal positioning docs, or unpublished research. That's what makes Knowledge Bases a defensible advantage.
- Call transcripts are your highest-value sourceProspects describe their pain points in their own words on sales and discovery calls. When you feed those transcripts into a Knowledge Base, your Workflows can pull real customer language into content automatically. You stop guessing what resonates and start writing with the exact phrases your buyers use.
- One Playbook query replaces hours of manual researchWhen a Playbook runs, it queries your Knowledge Base using semantic search and pulls back the most relevant chunks automatically. You stop copying and pasting from Google Docs into prompts and start letting your system surface the right context for each piece of content on its own.
- Schedule syncs so your Knowledge Bases never go staleYou can set up recurring data syncs on any connected source, like a Friday refresh for your site map or Google Drive folder. This keeps your Knowledge Bases current without anyone remembering to re-import. Your Workflows always pull from up-to-date material.
- Your published content is the starting point, not the finish lineA site map Knowledge Base indexes what you've already published, but the real advantage comes from importing what hasn't been published yet. Internal positioning docs, competitive research, and sales enablement materials give your Workflows access to context that no public search engine or LLM can find.
FAQs
A Knowledge Base is where you store your brand's unique context so every Workflow and agent in AirOps can access it automatically. Think of it as your content's memory: it holds documents, research, customer interviews, product docs, and other reference material, then surfaces the right information at the right time using semantic search. Unlike your Brand Kit, which controls how you sound, a Knowledge Base controls what your content knows.