Issue 002 · May 2026 · Private Beta

One memory. Every AI.

Notes, meetings, decisions, code snippets — drop in whatever shape your work takes. Cortex is the shared memory layer for Claude, ChatGPT, Cursor, and anything that speaks MCP.

No code? Skip the line.60-second survey · priority access + launch discount
Limited early-access spots/Works with Claude · ChatGPT · Cursor/Solo or team
“Same memory across Claude, ChatGPT, and Cursor. I haven’t re-explained myself to an AI in three weeks.”Tim, freelancer — Rubber Ducky Studio
· The breakthrough ·

Every AI. One brain.

MCP is an open standard. Every AI that speaks it can read your Cortex — Claude, ChatGPT, Cursor, Claude Code today. Capture once. Cite everywhere. Stop being the human copy-paste layer between your own tools.

Claudevia Cortex MCP
What did we agree with Jan?
search_brain("jan agreement")→ 3 neurons
Terms were agreed Tuesday — contract signed[2]. 4 follow-ups from the kickoff[3].
ChatGPTvia Cortex MCP
What did we agree with Jan?
search_brain("jan agreement")→ 3 neurons
Terms were agreed Tuesday — contract signed[2]. 4 follow-ups from the kickoff[3].
Cursorvia Cortex MCP
What did we agree with Jan?
search_brain("jan agreement")→ 3 neurons
Terms were agreed Tuesday — contract signed[2]. 4 follow-ups from the kickoff[3].
Your Cortex brain — shared memory3 neurons cited by every MCP client
1
PROJECT
Rubber Ducky redesign
Scope, timeline, and client budget
2
EMAIL
Thread with Jan — contract
Terms agreed Tue · PDF attached
3
NOTE
Voice memo — kickoff
4 follow-up tasks auto-extracted
Claude·ChatGPT·Cursor·Claude Code·+ any MCP client
Same answer. Same citations. Same memory — because every MCP client reads your Cortex.
· Anatomy of one memory ·

Pipe what you use in. Read it from any AI.

That’s the whole product. Connectors push your work into one brain. MCP exposes it to every AI you use. The model becomes replaceable; your memory doesn’t.

· Connectors push in ·
Gmail
Slack
Drive
Clipper
↓ pipes in ↓
Brain
one memory
↓ via MCP ↓
· MCP clients pull out ·
Claude
ChatGPT
Cursor
Claude Code
1
Capture

Gmail, Slack, Drive, the Clipper, file drops, voice notes. Each pipes into one brain — scoped to the right project.

2
Index

Cortex embeds in the source language, auto-tags, and links to people. Hybrid retrieval keeps recall sharp.

3
Read

Every MCP-speaking AI calls the brain mid-conversation. Same context whether you’re in Claude, ChatGPT, or Cursor.

· Inside Cortex ·

Capture once. Cite everywhere.

Four pieces that work together. Projects to slice your brain by client. The brain itself to hold what you know. Every AI tool you use to read it. Signals to keep it moving.

Projects = brain slices

One project per client.
The AI knows which.

Projects exist to divide brain knowledge — one per client, one per engagement, one per product. Set the active project; every connected AI follows. No more 600-word system prompts pasted into every new chat.

  • Switch the active project; Claude, ChatGPT, Cursor all follow
  • Per-client scope means client A never leaks into client B
  • Invite a teammate or client into a single scope
How project scoping works
cortex.ai / projects / acme-cloud
Acme Cloudactive
Multi-tenant ML inference · 87 neurons · 4 members
Reading scopeAcme Cloud
Claude DesktopCursorChatGPTClaude Code
MPDS
4 members2 signals · 1 running
Recent — this scope
DECISION2h
ADR-002 — Kafka for the job queue
Decided last week. Confluent Cloud, 3 brokers, replay enabled.
EMAILyesterday
Q3 budget approval — Anna confirmed €4,500/mo
Forwarded yesterday. Final terms in PDF.
CLIP4d
Substack: Why B2B funnels are broken
Saved from your browser. Tagged #marketing.
Other scopes
Lighthouse Labs41Marketing28Personal213
Switch scope · every connected AI follows
The Brain

Searchable by meaning,
in any language.

Notes, documents, quotes, embeds — one searchable memory. Save in Dutch, ask in English. Cortex embeds in the source language and matches across. Hybrid retrieval finds the right neuron even when the words don't match.

  • Hybrid retrieval — dense + sparse + reranked
  • Cross-language search out of the box
  • AI-tagged on capture; revisions + private working notes per neuron
See the Brain
cortex.ai / brain · scope: Acme Cloud
Ask your brain anything…⌘K
All87Notes34Decisions7Events12People7Quotes9Files18
DECISIONAcme Cloud2h
ADR-002 — Kafka for the job queue
Replay + back-pressure beat polling Postgres. Confluent Cloud, 3-broker.
Brain
#adr#infra
EMAILAcme Cloudyesterday
Q3 budget — Anna confirmed €4,500/mo
Forwarded yesterday from Anna at Acme. Final terms in attached PDF.
Gmail
#budget#q3
PERSON4d
Diego Hernández — PM · Acme Cloud
24 neurons · last seen in Q3 budget thread · owns billing v2
Brain
QUOTEMarketing1w
“Why B2B onboarding funnels break” — Lenny
Saved from browser via Clipper. Tagged automatically.
Clipper
#marketing
CHATAcme Cloud3d
Debugging flaky e2e tests — Claude
Saved conversation. Root cause + fix path captured for the team.
Claude
#debug#tests
EVENTAcme Cloudin 2d
Q2 planning offsite — Tue 14:30
Captured from voice memo. Attendees + agenda extracted.
Voice memo
#planning
Hybrid retrieval — dense + sparse + reranked87 neurons indexed
Prompt neurons

Save the prompt.
Invoke it by name.

Your hard-won prompts are knowledge too. Save a software-engineer review prompt, a security reviewer, a brand-voice copywriter — once, scoped to you, your team, or a project. Invoke from any AI: `use cortex software-engineer+security-reviewer review this codebase`. Combine two with `+`, add a request, and Cortex composes the rest.

  • Personal, team, or project-scoped — same access model as every other neuron
  • Combine slugs with `+` to merge prompts before your message
  • Six built-ins to start — engineer, security, copywriter, strategist, researcher, writer
See prompt neurons
cortex prompt — compose
Prompt neurons
software-engineerpersonal

Senior dev mindset — call out hidden invariants, prefer the smallest change…

security-reviewerbuilt-in

OWASP-aware review. Output structured findings as a JSON array a scanner can ingest.

invocation
use cortex software-engineer+security-reviewer review this codebase
composed body sent to model
One brain, every AI

Claude. ChatGPT. Cursor.
Same memory.

MCP is an open standard. Plug Claude in. Plug ChatGPT in. Plug Cursor in. They all read from — and write to — the same brain. Switch models monthly. Your context stays put.

  • Claude Desktop, Claude.ai, ChatGPT, Cursor, Claude Code today
  • Refresh-token MCP — Claude.ai stays connected for weeks, not hours
  • Search across brain, Gmail, and the live web in one unified call
See it in action
cortex MCP · 3 clients connected
ClaudeliveCursorChatGPTscope: Acme Cloud
What did we decide about the job queue, and who owns billing v2?
search_brain("job queue · billing")→ 4 neurons
cortex://recent→ scope Acme
Per ADR-002, you went with Kafka over polling Postgres for replay and back-pressure[1]. Billing v2 is owned by Diego Hernández (PM) with Priya on the ledger-model migration[2].
[1]ADR-002 · Kafka for the job queue[2]Person · Diego Hernández
Save this conversation to the brain.
save_to_brain(kind="chat", scope="Acme Cloud")✓ saved
Same brain, same answer, in any MCP client
Signals

Automate
the boring.

"Every Monday 09:00, draft the weekly update from shipped work." Signals run AI on a schedule with your brain as context, and land the result as a neuron — searchable, citable, ready to ship.

  • Cron, event, or webhook triggers — every 5 minutes to yearly
  • Recipes pre-fill the trigger and prompt; per-run detail to inspect
  • Actions: Slack with templates, webhooks, AI-generated neurons
See Signals
cortex.ai / signals
4 active·1 running nowNext run · 14m
Friday client status — Acme CloudFri 16:00running
Drafting from 18 neurons added this week · 4 decisions, 2 incidents…
New PR — summarise & fileOn webhookran
PR #482 “Switch billing v2 to ledger model” → 6-bullet summary
Wrote 1 NOTE → Acme Cloud
Renewal radar — 60 days outDaily 09:00ran
3 accounts within 60 days · QBR briefs queued for Diego, Sara, Jasper
Wrote 3 EVENTS → 3 scopes
Daily support sentimentEvery 4hpaused
Paused while we re-tune the recipe
RecipesPre-filled trigger + prompt
Weekly client updateRenewal QBRStand-up digestIncident postmortem
· Ask · Answer · Ship ·

Ask your brain anything. Get an answer from the AI you’re already in.

Plug Cortex into Claude, Cursor, ChatGPT, or any MCP client — then call your saved prompt neurons by name. Same /p slug from anywhere.

Claude Desktop
Claude Desktop
/p onboarding
prompts.resolve("onboarding") → composed body sent
## The mission
Acme Cloud — multi-tenant ML inference, two control-plane regions, Q2 priority is billing v2.
## Your stack
Go · Next.js · Postgres · Kafka · GKE.
## The team
Marcus (CTO) · Priya (staff) · Sam (SRE) · Diego (PM)…
## Your first week
Pair with Priya on inference path · ship a tiny PR by Friday.
Powered byonboardingproject
Cursor
Cursor
/p software-engineer+install-guide spin up a dev env
prompts.resolve("software-engineer+install-guide") → 2 prompts merged
// 1. Clone & bootstrap
git clone acme/cloud && pnpm install
// 2. Postgres (per ADR-002)
docker compose up postgres
// 3. Kafka for the job queue
docker compose up kafka redpanda
// 4. Run it
pnpm dev
Powered bysoftware-engineerbuilt ininstall-guideproject
ChatGPT
ChatGPT
use cortex org-chart who’s the PM?
use-cortex parser → 1 prompt + your trailing question
DH
Diego Hernández
Product Manager · Acme Cloud
Emaildiego@acme.cloud
Slack@diegoh
OwnsBilling v2 · Q2 roadmap
Powered byorg-chartteam
Claude Code
Claude Code
/p decisions
prompts.resolve("decisions") → composed body sent
ADR-001Postgres over MongoDB
ADR-002Kafka for the job queue
ADR-003Multi-region active-active
ADR-004gRPC for control plane
+ 6 more from sprint retros
Powered bydecisionsproject
Save it once as a prompt neuron. Invoke from any AI with /p slug or use cortex slug. Combine two with +.
· A day in Cortex ·

How a creative agency stops paying the freelancer tax when it has fifty employees instead of one.

Account teams burn 30–40% of a campaign cycle on coordination overhead — re-explaining clients to AI tools, re-reading old briefs, re-onboarding new joiners. Cortex makes one shared brain per client a default, not a heroic individual effort.

  1. Tuesday 09:30
    Client A, full team aligned
    Strategy, creative, and project management all open Claude with project Acme active. The AI gives every role the same grounded answer — no four contradictory drafts.

    One brain per client, shared by your team. The AI’s answers reference the same canonical neurons whether the strategist or the designer asks.

    Team brains →
  2. 11:00
    Brief lands; brain absorbs
    Client emails the brief. The drafts inbox suggests it as a project Acme neuron. Two clicks, in. Claude can now ground every campaign idea against the actual brief, not a paraphrase.

    Connector-driven inbox surfaces incoming context — Gmail today, Slack soon. Accept into a scope; the brain stays canonical, not someone’s notes app.

    File drop →
  3. Thursday 14:00
    Handoff without the briefing call
    Strategy hands creative the campaign. They open project Acme. Same brain. The handoff Loom is a search query, not a meeting.

    Scoped brains beat shared docs because the AI grounds in them automatically. Less meeting time; fewer dropped balls.

    Every AI reads your brain →
  4. Friday 16:00
    Status report writes itself
    A signal compiles the week’s decisions, asset deliveries, and open client questions per account. Account directors edit; the client gets a status update before anyone opens Asana.

    Signals run AI on a schedule with the project brain as context. Reports come from real activity, not memory.

    Signals →
  5. When someone leaves
    Knowledge stays
    Senior creative director moves on. Her brain didn’t walk out with her. Their replacement starts with the same context, not a wiki crawl.

    Every project has a shared brain that survives departures. Onboarding compresses; institutional memory becomes durable.

    Team brains →
Observed in the wild

Sara, account director — 22-person creative agency

· Solo or team ·

Run solo. Scale with a team.

Monday: client A. Tuesday: client B. Wednesday: an internal tool you barely remember. Cortex is built for that life — one project per client, every AI you use already knowing which. Same brain scales the same way when you grow into a team or invite a client into their own scoped slice.

Solo · €19 / month

One brain. Every client.

A project per client; the AI knows which one is in front of you. No more pasting the same 600-word system prompt into every new chat. Private by default — not even we read it.

  • Switch the active project; every AI follows within a second
  • Every MCP client reads from it — Claude, ChatGPT, Cursor, Claude Code
  • Browser Clipper covers 50+ web platforms straight into your brain
See the Brain
Team · €29 / seat

One brain per client. Shared by your team.

Agencies and in-house teams: scope the brain per client, per engagement, or per product. Onboarding stops being a six-month wiki crawl. New teammates ramp by searching.

  • Project- and team-scoped memory with audit log
  • Share a project with one team, several teams, or the whole org
  • Add a teammate; their first day looks like their fourth
See Team brains
· Integrations ·

Pipe the tools you already use into one brain.

Live today: Gmail, Slack, Drive, the Browser Clipper for 50+ web platforms, plus Anthropic / OpenAI / OpenRouter / Ollama on the model side. Every connector is also an MCP tool — connect once, every AI you use can call it.

GmailSlackGoogle DriveBrowser ClipperAnthropicOpenAIOpenRouterOllamaGitHubQ3LinearQ3JiraQ3NotionQ3OutlookQ3iCloudQ3FigmaQ3GmailSlackGoogle DriveBrowser ClipperAnthropicOpenAIOpenRouterOllamaGitHubQ3LinearQ3JiraQ3NotionQ3OutlookQ3iCloudQ3FigmaQ3
· Who uses it ·

Built for one freelancer. Useful to four more.

Cortex started as one operator’s planning brain. The same shape fits anyone juggling clients, projects, or roles.

Five sharp personas instead of twenty vague ones.
· Pricing ·
Free during private beta

After launch: €19/month solo, €29/seat for teams. Everyone who joins during the private beta gets three months free and a year of price-lock.

Claim early access
· Apply for early access ·

One memory. Every AI you use plugs in.

Private beta. Limited spots. Redeem your code to jump in — or grab a waitlist seat at the bottom of the page.