Keep Cortex local-first. Cloud sync should remain optional, encrypted, and explicitly enabled.
Give every AI tool your memory and taste.
Cortex turns your notes, decisions, preferences, and AI history into one reviewed, cited layer on your Mac—so every assistant remembers what matters and makes choices that feel like you.
Approve its memory. Shape its taste.
Direct, warm, technically credible.
Ready to reviewVerify the release update feed.
Ready to reviewRemember once. Bring your taste everywhere.
Memory carries what you know. Taste carries how you choose. Cortex makes both visible, reviewable, and portable.
Connect your sources
Import notes, AI chat exports, and the folders where your work already lives.
Files · exports · local foldersReview memory and taste
Approve facts, decisions, preferences, and style signals before Cortex trusts them.
Human-reviewed by defaultApply it anywhere
Give every approved AI tool the context and taste it needs to work the way you would.
Claude · ChatGPT · Cursor · MCPYour domain has a way of thinking.Shape Cortex around it.
Start with the same private, reviewed vault—then decide what Cortex should remember, which preferences it should follow, and how it should make subjective choices for the work you do.
Give coding agents the decisions and engineering taste behind the code—not another undifferentiated transcript.
Architecture taste and decisions
Let Cursor or Claude retrieve why a system is shaped this way, which patterns you prefer, and what made the final choice win.
Incident continuity layer
Carry symptoms, experiments, fixes, and follow-ups across an incident so a fresh session can continue the investigation without starting over.
Cross-tool project handoff
Move from ChatGPT to Cursor to Claude with the active branch, conventions, unresolved questions, and next actions still intact.
Turn a growing source library into reviewed evidence—and preserve your judgment about what is rigorous, relevant, and worth pursuing.
Living literature map
Keep approved claims, methods, disagreements, and source passages connected as the reading list grows.
Experiment ledger
Remember hypotheses, parameters, unexpected results, and next steps across runs without flattening the history into a summary.
Cited synthesis partner
Assemble only reviewed evidence into outlines and drafts, with every recalled idea linked back to its original source.
Keep the small decisions, product taste, and recurring signals that disappear between customer calls, planning, and fundraising.
Customer signal memory
Connect interviews, support threads, and product notes so recurring pain points can be recalled with the exact conversations behind them.
Company decision room
Preserve product, hiring, and operating decisions with owners, rationale, and revisit dates instead of losing them in chat.
Fundraising continuity
Keep metrics, milestones, narrative taste, and investor questions consistent across decks, emails, and follow-up conversations.
Build a private learning model around what you understand, what still feels unclear, and which explanations fit the way you learn.
Personal study model
Let an AI tutor remember mastered concepts, recurring misconceptions, and explanations that clicked—after you review them.
Course knowledge vault
Connect lectures, readings, and assignments by concept so every new session starts with the right course context.
Research coach
Keep citations, promising questions, and incomplete arguments available across writing tools without inventing missing sources.
Give creative tools continuity without letting the model quietly rewrite the taste, voice, or rules of the project.
Voice and taste vault
Store approved references, tone rules, preferred language, and the things a project should never sound like.
Long-project continuity
Keep characters, scenes, design decisions, and unresolved threads coherent across weeks of work and many AI sessions.
Reference recall
Find not only the reference you saved, but why it mattered and which part of the project it was meant to influence.
Judgment you can trace back.
Cortex uses only approved facts, preferences, and decision patterns, explains why they matched, and refuses to invent context or taste when the evidence is not there.
Read about CMPHow should the launch announcement sound?
Lead with the local-first promise. Keep the voice direct, warm, and technically credible—and avoid inflated AI claims.
Your memory and taste stay yours.
Cortex is designed for private context and subjective preferences: local storage, explicit approval, visible sources, and permissions you can revoke.
Local by default
Your context, preferences, and taste stay on your Mac and work offline.
Reviewed, not assumed
Nothing becomes trusted memory or taste until you approve it.
Cited or abstained
Every recalled fact or applied preference points to a source—or Cortex stays quiet.
Permissioned per tool
Read, save, export, and repair are separate, revocable controls.
Your memory and taste, in a vault you can open.
Facts, decisions, preferences, and style live in plain Markdown with a rebuildable index—all stored on your Mac and portable like an Obsidian vault.
- ⌄decisions14
- ⌄taste-and-style8
- ⌄open-loops6
- ⌄sources23
Open source, all the way down.
Inspect the SwiftUI app, the local FastAPI engine, the vault format, how taste is applied, the retrieval quality gate, and every permission boundary.
# native macOS app
./macos/build.sh
# local engine + tests
python3 -m pytest backend/tests
# retrieval quality gate
python3 scripts/retrieval_eval.pyYour AI should know what you know—and how you choose.
Memory and taste, private on your Mac, reviewed by you, and open source for everyone.