Cortex Flagship build · Self-hosted

Cortex OI

A working example of operational intelligence built around a real business.

Cortex turns operating documents and industry signals into cited recommendations and a decision history your team can revisit. It shows the kind of focused system an embedded engagement can produce.

Discuss a build

A Cortex-focused build fits inside the same $2,500/month embedded engagement.

Decision archive Resolved · Q3

Should we add commercial battery storage to our offering this year?

Recommendation Proceed with conditions
Confidence Medium-high
Risk flags 2 flagged
Dissent BESS margins thin without IRA credits
Sources Q3 sales notes OEM partner thread Regional market scan

The Problem

Important decisions rarely arrive with organized context

Your business runs on knowledge scattered across folders, inboxes, and undocumented meetings. When the next big call comes (hire or outsource, raise or bootstrap, expand or focus), you're working from fragments.

Generic AI chatbots do not know your operating history. Cortex is an example of a system built around it: your documents, your prior decisions, cited sources, and explicit dissent.

Trust

Built so you stay in control

Your data never leaves

Cortex runs in your environment, with documents, KG, and decisions on infrastructure you control. Nothing is mined for training.

Every answer cites sources

Every output points back to the documents, KG nodes, and prior decisions that grounded it, so there's no black box to argue with.

You decide. Cortex recommends

Cortex is the intelligence layer, not the agency layer: it surfaces options and flags risks, and you make the call.

A Real Artifact

What an output actually looks like

One deliberation, condensed. This is what lands in your decision archive when the council weighs in: every persona named, every claim sourced, dissent on the record.

Question
Should we add commercial battery storage to our offering this year?
Recommendation Proceed with conditions
Confidence Moderate
S
Strategist Yes

Three commercial inquiries last quarter and an OEM offering co-marketing dollars. Skipping the segment hands the channel to a regional competitor.

Sources: Q3 sales notes · OEM partner thread · regional market scan
P
Pragmatist Yes, staged

Bid two pilot projects in the next 90 days before committing to crew hires. Install team is at 78% capacity; commercial code differences need recertification on two staff.

Sources: weekly capacity reports · project pipeline · certification log
K
Skeptic Conditional · 3 risks

Margin compression as utility-scale players enter the segment. Cash flow drag from 90-day commercial payment terms. Crew skill gap: commercial code is materially different from residential.

Sources: 2024 industry margin report · AR aging history · install team certifications
L
Lean Black Belt Smaller cut available

Start with battery-only retrofits on existing solar customers. Lower install complexity, established trust, ~12-week payoff per project. Half the capital exposure of a full commercial play.

Sources: customer database · historic install payback data
Dissent on the record

Skeptic recommends a 90-day pilot before any crew expansion. Strategist disagrees, citing the 6-month commercial sales cycle and OEM co-marketing window. Both views logged.

Revisitation triggers

OEM rebate program changes · install team capacity exceeds 90% · first two commercial bids closed

The Difference

Why this isn't just ChatGPT

Cortex isn't a replacement for ChatGPT or Claude. It's everything raw LLM use is missing for actual business decisions: your context, real citations, multi-perspective pushback, and memory that compounds.

Raw ChatGPT or Claude
Cortex OI
Knows your business
Generic web
Ingests your documents into a structured knowledge graph
Stays current
Frozen at cutoff
Continuously scouts and ingests industry and market intelligence
Pushes back
Sycophantic
Skeptic on every recommendation
Tone
Validates you
Truthful and candid, like a real board. Optimized for accuracy.
Citations
Hallucinated
Every answer cites real KG nodes
Memory
Stateless
Remembers what you decided and reopens it when conditions change
Improves over time
Static
Calibration loop tunes from your corrections

How It Works

How Cortex OI thinks

Your documents and live industry feeds become structured knowledge. Your council deliberates. You get one clear answer with confidence, citations, and dissent on the record.

Your Documents

PDFs, decks, transcripts, financials. All ingested.

ingest

Industry Feeds

News, regulatory updates, competitor signals. Continuously scouted into the same KG.

scout

Knowledge Graph

Structured, searchable, always current.

query

Advisory Council

Strategist, Pragmatist, Skeptic, Lean Black Belt, plus a Domain expert tuned to your business. They deliberate. They disagree. Dissent stays on the record.

synthesize

Cortex OI

One clear answer. Confidence-scored, sources cited, dissent on the record.

YOUR DOCUMENTS Financials Strategy PDFs, decks, transcripts INDUSTRY FEEDS News Regulatory Competitors Live, autonomously scouted ingest scout KNOWLEDGE GRAPH Vendor Deal Risk Decision Customer Structured. Always current. query ADVISORY COUNCIL S Strategist P Pragmatist K Skeptic L Lean + a Domain expert tuned to your business Deliberate. Disagree. Document dissent. synthesize CORTEX OI Recommendation High Risk 2 flags Sources 4 refs Dissent noted One voice. Full picture. calibration loop
Calibration loop
Your corrections compound over time. Cortex gets sharper the longer you use it.
Decisions on the record
Recommendations land with reasoning, citations, and dissent. Cortex reopens them when the underlying conditions change.

The Council

A real board, not a chorus

Cortex agents behave like a trusted council. Truthful, candid, and on the record: lead with the counterargument when there is one, give explicit confidence (high, moderate, low, unknown), and tell you when you're wrong instead of telling you what you want to hear. Generalists on every decision, plus specialists tuned to your business when the question demands depth.

S

Strategist

Holds the long view. Competitive position, optionality, the story you'll tell three years from now.

P

Pragmatist

Holds execution. What ships, what blocks ship, the real sequence of next moves.

K

Skeptic

Holds risk. What could go wrong, which assumptions are weakest, what would change the call.

L

Lean Black Belt

Holds capital efficiency and process discipline. The smallest version that proves the bet, what to cut if budget halves, where the waste lives.

+ Specialist agents tailored to your business

Setup includes a Domain agent seeded from your documents. Add Legal, Finance, Industry, or any other lens your decisions warrant. Council depth scales with the question: light for routine calls, deep when the stakes (or the budget) call for it.

Example scope

What a Cortex-focused build can include

Cortex is one example, not a separate software subscription. Within an embedded engagement, a build like this can be deployed in your environment and improved as it proves useful.

First working build
  • Deployed in your environment (your servers, your data)
  • Document ingestion: PDFs, decks, transcripts, markdown, plain text
  • Industry intelligence scouts: news, regulatory, competitor signals tuned to your market
  • Knowledge graph seeded from your documents and live feeds
  • Generalist council (Strategist, Pragmatist, Skeptic, Lean Black Belt)
  • Domain persona authored for your business
  • Decision archive initialized
  • Team onboarding and authoring guide
While the engagement continues
  • Monthly tuning of council personas and the Domain agent
  • KG maintenance: new docs ingested, stale nodes pruned
  • Calibration loop processed (your corrections compound over time)
  • Decision archive review: revisitation triggers surfaced
  • Direct line for product questions and ad-hoc advisory
  • Hosted AI usage is billed through your own key; actual cost depends on model and volume

Want Cortex deployed in your business, or something custom-built that solves a different operational problem? See how engagements work.

Questions

Frequently Asked Questions

In your environment. Self-hosted on infrastructure you control: your laptop, an office server, or your cloud account. Nothing leaves.

Cortex is built so the answer doesn't depend on me. Your KG and decisions are plain markdown. Your conversation history is SQLite. Your model runs on your own Anthropic or Ollama key. The whole deployment lives in your environment. If I step away from the project, you keep the data, the archive, and a working system. Source code escrow available on request.

You bring your own Claude or local-Ollama key, so hosted-model usage is visible in your own account rather than buried in a SaaS line item. Actual cost depends on the model, usage, and depth of deliberation. A local Ollama setup avoids hosted-model token charges.

Default is Claude (Sonnet for personas, Haiku for routing) using your own API key, so you control rate limits, costs, and version. Local Ollama models supported for full air-gap.

No. Cortex uses your data to ground responses through retrieval, not to train. Corrections feed a calibration loop that lives only in your deployment.

Yes. Multi-user from day one with admin and member roles, shared KG, shared decision archive. Built for teams of 1 to 20.

One to two weeks. Week one: deploy, ingest, seed KG, author Domain persona. Week two: tune, train your team. Then it's live.

Every output ships with confidence, citations, and dissent. When something's wrong, you correct it in-line. The correction tunes future answers so the same mistake is less likely.

Everything is yours. KG and decisions are plain markdown. Conversation history is SQLite. Cancel any time and keep your data plus a frozen snapshot of the system.

Pre-revenue startups still finding product-market fit. Teams looking for someone to make the call for them (Cortex argues, you decide). Anyone unwilling to ingest documents or write down how their business actually runs. If you want a chatbot that sounds confident no matter what, this is the wrong tool.

Could a system like this help your business?

A 15-minute call is enough to decide whether a Cortex-focused build fits your problem, or whether a different operational tool would serve you better. Both happen through the same monthly engagement; how engagements work has more detail.

Book a 15-min call

Tell me what decisions you're working on, and I'll be honest about whether Cortex is the right tool or whether something else would serve you better.