Cortex · Brand memory for AI
Cortex turns your scattered brand knowledge — voice rules, frameworks, case studies, personas, portfolio, reviews — into structured memory every AI agent in your stack can query. One schema. One source of truth. No more PDFs.
The problem
Every new AI tool your team adopts starts from zero. Voice guides, case studies, personas, and proof points scatter across Notion, Google Docs, slide decks, and style guides — readable by humans, useless to agents. So every model makes it up.
Notion pages, PDFs, Slack threads, Figma files. No agent knows where to look or what to trust.
Style guides are written for humans. A model reads "be conversational but authoritative" and invents its own rules.
Case studies hallucinate. Customer quotes get fabricated. Numbers appear from nowhere.
Sales, content, and support each train different AI prompts. Every tool speaks a slightly different brand.
What Cortex stores
Every field of brand knowledge is typed, versioned, and queryable through a single protocol. Agents know exactly what to ask for — and exactly what they'll get back.
Positioning, boilerplate, one-liners, naming conventions.
Tone rules, authority phrases, banned words, cadence preferences — as structured data, not prose.
Colour tokens, type scale, logo usage. Ready for any agent that renders visuals.
Your proprietary methodologies with permalinks, so every agent can reference and link them correctly.
Structured ICP and buying-committee profiles with pains, gains, and objections.
What your team has discovered about messaging, positioning, and market — captured as reusable memory.
Real client outcomes with hard numbers, quotes, and linkable URLs. No invented proof.
Work samples with metadata — industry, outcome, asset URLs — searchable by agents.
Customer quotes with attribution, source, and date. Audit-ready.
Demo clips, testimonials, and explainers indexed for retrieval by topic.
The platform
Each layer is designed for a specific job an agent needs to do. Together they turn brand knowledge into something a model can reason with.
Voice rules, authority phrases, frameworks, and proof live as structured data — not a style-guide PDF the model glances at. When an agent asks "how do we talk about pricing", it gets an answer, not a guess.
Case studies, portfolio, reviews, and videos are stored with outcomes, attribution, and links. Agents can search them by topic, client, or industry — so proof gets cited by name, never invented.
Cortex speaks Model Context Protocol — Anthropic's open standard for agent tools. Claude Desktop, Claude Code, Orbit, and any MCP-compatible client can query brand knowledge through the same interface.
Every brand gets its own namespace inside Cortex. Data never crosses boundaries — voice rules for one brand never leak into another brand's agent context.
Cortex syncs to a local wiki cache, so agents read brand knowledge from disk on hot paths and fall back to the live MCP only when the cache is stale. Brand context loads once per workflow — not once per call.
Who it's for
Orbit uses Cortex as its brand brain. Every article it writes pulls voice rules, frameworks, and proof from a single source — no drift, no invented stats, no stylistic hallucination.
Sales agents, support agents, research agents — every Claude-based tool in your stack can query Cortex and answer with the same facts, the same voice, and the same proof.
Your brand guide is never "done". Cortex keeps it living — editable, versioned, queryable by every tool your team uses. The PDF era is over.
Cortex + Orbit
Orbit — our AI content operations product — is built on top of Cortex. Every article Orbit writes pulls voice, frameworks, and proof from Cortex. If you want to see Cortex in action at full tilt, Orbit is the shortest path.
Explore OrbitUnder the hood
Every design choice assumes an AI agent is the primary consumer — not a brand manager reading a PDF.
HTTP and stdio transports, typed tool definitions, standard auth. Works with any Model Context Protocol client.
Every brand attribute is a typed field. Agents know exactly what they can query and what shape the response takes.
Distributed SQLite means Cortex is globally replicated, read-anywhere, writeable from any region.
Local markdown cache for token-efficient agent workflows. Live Cortex is the source; the wiki is the hot path.
Every change to voice, proof, or framework is versioned and audit-trailed. Roll back or attribute with ease.
Multi-brand by default. Per-brand access, per-brand audit logs, zero cross-brand data bleed.
Early signal
We had voice rules scattered across five Notion pages and a 40-page PDF. Cortex turned them into something every Claude agent in our stack actually respects.
Every agency promised a 'single source of truth'. Cortex is the first one that every AI tool we use can actually read.
Case studies cited by name, with real numbers, auto-linked to the source — that alone made our content defensible.
Questions
Cortex is an MCP (Model Context Protocol) server that stores and serves a company's brand knowledge as structured data. Anything an AI agent needs to represent your brand — voice rules, frameworks, personas, case studies, portfolio, reviews — lives in Cortex with a typed schema. Any MCP-compatible agent can query it.
Notion and PDFs are written for humans. Cortex is written for agents. Voice rules are structured fields, not paragraphs. Frameworks have permalinks so they auto-link when referenced. Authority phrases are detectable tokens a humanisation pass can enforce. A Notion page full of good intentions is useless to a model; Cortex is the schema a model can follow.
No. Cortex is a standalone product. It works with Claude Desktop, Claude Code, Orbit, or any custom agent that speaks Model Context Protocol. Orbit happens to be the first product we built on top of it, but Cortex is useful anywhere an AI agent needs brand context.
Your brand team, content team, and ops team — through a simple editing UI. Changes are versioned and audit-trailed, so you can see who edited what and roll back if needed. Agents pick up changes on the next call.
Per-brand namespaces with strict isolation. Agencies and holding companies run dozens of brands off one Cortex instance, and no brand ever sees another brand's voice rules, proof, or personas.
Yes. Cortex ships as an MCP server you can run on your infrastructure, with managed hosting available for teams that prefer it. Enterprise customers can run it on-premise or in a VPC with custom auth.
Typed responses. Ask for company info and you get a structured object with name, positioning, brand voice, colours, and frameworks. Ask for case studies and you get records with client, outcomes, quotes, and source URLs. The schema is discoverable — agents introspect what's available.
Ready?
We'll import your existing brand assets — voice guide, case studies, portfolio, reviews — and stand up a Cortex namespace for your brand. Usually within a week.