Cortex / Knowledge
Self-hosted · Privacy-first · Citation-required

Make 12 years of work product instantly findable.

Cortex Knowledge is a self-hosted document Q&A system for small professional services firms. Drop in your PDFs, DOCXs, and notes. Ask questions in plain English. Every answer cites its source.

The problem
"We have 12 years of work product nobody can find. The senior people remember where things are. The junior people give up and ask the seniors. We waste hours per day."

Generic search doesn't synthesize, so you still have to open every result. ChatGPT can synthesize but can't be trusted with client data and won't cite sources. Both fail this use case for the same reason: small professional services firms need both precision and privacy, and neither alone is enough.

How a query works

Watch a real query in action.

cortex — knowledge query
$ cortex query "What was our position in the Henderson matter?"
[14:23:01] Searching 4,217 indexed documents...
[14:23:02] Top matches: smith-v-henderson-brief.pdf, henderson-strategy-memo.docx
[14:23:03] Synthesizing answer with citations...
[14:23:04] Answer ready.
Our position was that the prior agreement [1] did not extend to
consequential damages because the carve-out clause [2] specifically
excluded such recovery.
[1] smith-v-henderson-brief.pdf, p.7
[2] henderson-strategy-memo.docx, p.3

Click any citation to open the source document at the cited passage.

Architecture

Three steps, end to end.

01

Drop files in

PDF, DOCX, TXT, Markdown. Watched folder or web upload. Auto-indexed.

02

Ask in plain English

RAG-powered retrieval against your vectorized library. Local embeddings by default.

03

Get cited answers

Every claim links to source doc and page. Click through to verify.

Runs in Docker on your hardware. Postgres + Qdrant for storage; embeddings generated locally via Ollama.

For answer synthesis, you choose the LLM mode at install: Local Ollama keeps everything on your hardware. BYO Anthropic API key sends only the cited chunks to Anthropic under your own contract. Our proxy routes through our infrastructure to Anthropic with per-token billing.

Your full document corpus never leaves your firm regardless of mode. Retrieval sends only the relevant 5–10 chunks per query.

Why this way

Self-hosted is the differentiator.

Cortex Knowledge ChatGPT / Claude Microsoft Copilot Glean / Hebbia
Where documents are stored Your infrastructure Vendor Vendor Vendor
Who controls the LLM You (local or BYO) Vendor Vendor Vendor
Per-seat cost None $20/user/mo $30/user/mo $50K–$200K/yr
Right-sized for 5–30 person firms Yes N/A Bloated Enterprise only
v1 features

What ships in the first release.

  • Document ingestion: PDF, DOCX, TXT, Markdown
  • Auto-indexing: new, changed, and removed files reflected automatically
  • Citation-required answers with click-through to source passages
  • Local user accounts (admin + members)
  • Three LLM modes: bring your own Anthropic API key, run on local Ollama, or use our proxy
  • Admin dashboard: doc count, index status, users, query log, usage stats
  • In-app updates from container registry
  • Single-command backup and restore
Pricing

Flat install fee plus optional managed support.

Concrete numbers vary by firm size and LLM mode. Contact for a quote during your demo.

Ready to see it on your documents?

Book a 30-minute demo. I'll show you the system running against a sample legal/accounting corpus, then we'll talk about what installation looks like for your firm.