The Knowledge Base

Your Documents, Organized

The knowledge base is the foundation of everything Pilot does. It's where your documents live after you upload them, and it's the only source material Pilot draws from when writing content.

When you upload a document — a PDF, a Word file, a markdown page, an HTML export — Pilot reads it, extracts the content, and maps it into a structured representation organized by topic. Not a folder. Not a search index. A knowledge map that understands what your documents are about and how they relate to each other.

What You Can Upload

Pilot accepts the formats your documents already live in:

• PDF — research papers, reports, whitepapers, exported presentations
• Word documents — memos, proposals, drafts, client deliverables
• Markdown — blog posts, documentation, notes
• HTML — web pages, exported articles, email archives
• Plain text — transcripts, notes, raw content

There's no required structure. Pilot reads the content regardless of how it's formatted. A polished annual report and a rough internal memo are equally valid source documents — Pilot extracts the knowledge from both.

What Happens After Upload

Upload is fast — files are accepted in under a second. Processing happens in the background. For each document, Pilot:

Reads the full text. Not just keywords or headings — the entire document, with context preserved.

Identifies topics. Pilot determines what subjects the document covers. A single document might cover multiple topics. A research paper on residential construction might touch fire safety, building codes, insulation standards, and labor economics — Pilot maps all of them.

Maps relationships. Documents don't exist in isolation. A report on OSHA compliance standards relates to a paper on fall protection equipment. A market analysis connects to a pricing strategy memo. Pilot identifies these connections and builds a web of relationships across your entire collection.

Indexes for retrieval. The document's content becomes available as source material for content generation, with enough structural understanding that Pilot can draw from the right sections of the right documents when writing about a specific topic.

How Topics Emerge

You don't create topics manually. They emerge from your documents.

Upload fifty articles about building science, and topics like "thermal bridging," "moisture management," "energy modeling," and "code compliance" appear because your documents discuss those subjects. Upload a collection of financial analyses, and topics like "revenue forecasting," "risk assessment," and "market positioning" surface.

Topics aren't rigid categories. They're clusters of related knowledge that Pilot identifies across your document collection. As you upload more documents, existing topics deepen (more source material on the same subject) and new topics appear (coverage of subjects you hadn't uploaded about before).

This is the behavior described in David Habib's Latent Vector as the activation of latent knowledge — information that exists in your documents but remains dormant until a system is built to surface it. Your organization's expertise on thermal bridging exists across a dozen different reports. Until those reports are in the knowledge base and the topic connections are mapped, that expertise is invisible. Pilot makes it visible — and writable.

A Living System

The knowledge base isn't a one-time import. It grows and changes as your organization produces new material.

Add documents, get new topics. When you upload a set of documents on a subject your collection didn't previously cover, a new topic cluster appears. New content opportunities surface through Pilot Lights.

Deepen existing topics. Upload more material on a subject you already cover, and Pilot has richer source material to draw from. Articles written from deeper knowledge are more substantive, with more sources to cite and more perspectives to synthesize.

Retire outdated material. When a document is no longer accurate or relevant, archive it. The knowledge base updates, and future content won't draw from outdated sources. Pilot doesn't cite material you've removed.

Watch it grow. The Pilot console shows your knowledge base's shape — which topics are well-covered, where gaps exist, and how documents connect. This visibility turns your knowledge base from an invisible asset into a manageable one.

What the Knowledge Base Is Not

It's not the internet. Pilot doesn't search the web. It doesn't supplement your documents with outside sources. If it's not in your knowledge base, Pilot doesn't write about it. This is a deliberate constraint — the value of Pilot's output depends on the integrity of its sources, and those sources are exclusively yours.

It's not a database you query. You don't write SQL against it or browse it like a file manager. The knowledge base is a backend system that Pilot uses when generating content. You interact with it by uploading documents and reviewing the topics and connections that emerge.

It's not a one-to-one map. One document doesn't produce one article. The knowledge base synthesizes across documents. An article about fall protection might draw from six different research papers, a regulatory document, and an industry report — all present in your knowledge base, all cited in the output.

Size and Scale

The knowledge base works with ten documents or ten thousand. The operational profile changes — a knowledge base with a thousand documents has richer topic coverage and more source material for synthesis — but the mechanics are identical.

Larger knowledge bases produce better content because Pilot has more material to draw from, more connections to identify, and more sources to cite. But even a small, focused collection produces useful output. A consulting firm that uploads its twenty best whitepapers gets articles that synthesize those twenty documents' worth of expertise. Scale comes when it comes.

Getting Started

To see how the knowledge base fits into Pilot's workflow, read How Pilot Works. To upload your first documents and watch topics emerge, follow the Getting Started guide.