What is Cortex?
What Cortex is
Cortex is our shared content and knowledge platform: one home for the official records, one map of how things connect, and search by meaning in long text and transcripts + plus admin and APIs — so we’re not rebuilding the same foundation in every project.
It keeps who said what, which companies and people matter, and how things relate organized. The system splits that work across three specialized parts; each part does a job the others are not meant to replace.
Three parts, three questions
Think of it as: records, connections, and “find similar wording”
| Part | Plain question | What it does for us |
|---|---|---|
| Document database | What are the things we care about? | The official profiles — companies, people, videos, taxonomies, and the fields you edit in admin. If it’s “the row we trust,” it lives here. |
| Graph database | How do those things link to each other? | The map of relationships — who invested in whom, who works where, what mentions what, playlists, and how content is classified in the graph. Good for questions like “who’s connected to whom?” |
| Vector database (Pinecone — research) | Which passages sound like this topic? | Semantic search over transcripts — “find moments like this” in video text, pointing back to the right episode. Not a replacement for the document store or the relationship map. |
One example each
- Document database: Someone updates Circle in admin — name, website, location. That’s the single profile the site and tools use.
- Graph database: You ask “which companies did this investor fund?” — that’s relationships between entities, not “search by similar words.”
- Vector database: A user searches “regulatory pressure on stablecoin reserves” — we find the most relevant transcript snippets and jump to the right video (and from there to people and companies if needed).
Why not one giant database for everything?
- The vector database (search-by-meaning) is optimized for huge piles of text fingerprints; it isn’t a full profile store or a relationship map.
- Profiles (what each thing is) get messy if we try to cram every relationship into the same place we edit companies and videos.
- Connections (who links to whom) are the wrong place to park massive text search indexes — we keep search beside the rest, not instead of it.
On top of that: Admin + APIs — one place to manage data and feed products, instead of rebuilding a mini backend for every initiative.
Why one Cortex beats many separate projects
Without something like this, teams often end up with:
- Different spreadsheets and apps with different versions of the same company or person
- No easy way to answer “who’s connected to whom?” across the business
- The same import and admin work rebuilt for every product
- Security and permissions harder to keep consistent
With Cortex as the shared layer:
- One place to trust — Editorial, product, and internal tools can align on the same facts.
- Relationships matter — We model how A relates to B, not only flat lists.
- One pipeline to care for — Seeding data, syncing to the graph, and APIs live in one place.
- Faster to ship new ideas — New surfaces can plug in instead of starting from zero.
- Room to grow — Categories, tags, and enrichment can evolve in one model instead of patching five systems.
One-liner for a meeting
Cortex is our shared content and knowledge platform: one home for the official records, one map of how things connect, and (when we turn it on) search by meaning in transcripts — plus admin and APIs — so we’re not rebuilding the same foundation in every project.
Suggested slide outline
- Problem: Scattered data and duplicate systems.
- What: Cortex = records + relationship map + (later) semantic search; each layer answers a different question.
- Why us: One source of truth, real questions about connections, one team maintaining the flow.
- Outcome: Easier to extend, easier to trust, better for features that span products.