Public Beta v1.5.1

One memory across every Claude surface.

Claude remembers now — just not in one place. Cortex makes it one memory.

Cortex is the control panel for what Claude knows — your memories, agents, skills, and projects, kept in sync across Claude Code, Claude Desktop, and Cowork. It runs on your Mac, and your data stays there.

Free · Runs on your Mac · Works in Claude Code, Claude Desktop & Cowork

Local-first. Runs on your Mac, no account.
Additive-only sync. Never deletes your files.
AI is optional. On your own Claude subscription.
Open self-host path for cross-device.

Claude remembers. Just not in one place.

Claude has memory now — but it's split across surfaces and kept by hand. Claude Code remembers one thing, Claude Desktop another, your phone a third. Move machines and you start over. The preferences you taught it last week live in a file you're quietly maintaining yourself.

Cortex is the shared memory and control layer underneath all of it. One place for your memories, agents, skills, and projects — captured as you work, kept in sync everywhere Claude runs, and stored on your Mac. You stop being the thing that remembers.

Six Layers of Memory

Not all memory is the same.

"Based on how brains actually handle memory — sensory at the surface, long-term at the base."

Cortex organizes what Claude remembers into six layers — from raw session signal at the surface, down to everything consolidated for the long haul.

Ephemeral → permanent

Sensory memory

The raw stream of this session — files touched, commands run, messages exchanged. Most of it disappears in seconds.

Working memory

What Claude is actively thinking with right now. A small handful of things held in mind at once.

Episodic memory

Specific moments from your recent work. Corrections you made. Decisions. Bugs you fixed.

Semantic memory

The rules and facts that cross every project. 'This team uses conventional commits.' Cross-project truth.

Procedural memory

Skills and workflows. The how, not the what. Built up from sequences that keep working.

Long-term memory

Everything consolidated for the long haul. Pulled back when you need it. Never the noise.

Unified surfaces

Your agents, skills, projects, and memory — in sync everywhere.

Cortex syncs both ways with Claude Code's own files and gives Claude Desktop and Cowork live tools. Build something once; it shows up wherever you work.

Claude Code

  • Agents & skills (~/.claude/…)
  • Project memory & session capture
  • Reads Cortex at the start of every session
Cortex

Desktop & Cowork

  • Live tools & the same memory, in the app
  • Pick up exactly where Code left off
  • Self-host · coming claude.ai & mobile via Connector

Create an agent in Cortex, use it in Claude Code seconds later — and the other way around.

Sync is additive-only: it can add and update, but it can never destroy. If a file goes missing, Cortex flags it with one-click restore; if the same thing is edited in two places, a Finder-style conflict sheet leads with a plain diff and lets you choose. Nothing is overwritten behind your back.

Over-remember, actively forget

Remembers everything. Forgets like you do.

Most memory tools turn into junk drawers. Cortex captures every session, lets what you stop using fade on a forgetting curve, and hardens what keeps mattering into durable rules — so Claude recalls the signal, not the noise. Nothing is ever silently deleted; weak memories just cool into an archive you can restore.

Step 1

Capture — every session, automatically.

When a Claude Code session ends, Cortex saves the substantive moments — decisions, corrections, preferences — as episodic memories. No "remember this" required. Capped, de-duplicated, and off with a single switch.

Step 2

Decay — the noise fades.

Memories you never reuse lose strength on an Ebbinghaus curve and settle into a cold archive — out of the way, never gone. Restore any of them anytime.

Step 3

Consolidate — patterns become rules.

What recurs across sessions hardens into durable semantic rules Claude can lean on — the difference between a pile of notes and something Claude actually knows.

Retrieval

Cortex knows what's relevant before Claude asks.

When you send a message, Cortex scans all six memory layers and surfaces only what matters — handing Claude the right context at the right moment.

Claude Cortex
DEVELOPER
What's our deploy process for Hopscotch?
• Cortex retrieved 3 memories
Procedural
Semantic
Episodic
Run scripts/release-testflight.sh — it handles the build bump, archive, and upload in one step. Commit with feat: or fix: prefix so the changelog reads cleanly. Last deploy was build 5 to TestFlight.

How It Works

1

Capture

Just work. When a session ends, Cortex catches the decisions and corrections that mattered — no save button, no "remember this."

2

Organize

It files each memory into the right layer — episodic, semantic, procedural — and lets the rest fade.

3

Surface

Next session, Claude already knows. No re-explaining. The context is there before you type.

The first session is the hardest one you'll have. After that, Claude has muscle memory.

Who Cortex Is For

Claude Chat User

You chat with Claude daily — desktop, phone, web. Cortex makes sure Claude remembers what you've been working on.

Mental Health Therapist

You see dozens of clients. Cortex holds case notes, frameworks, and boundaries so Claude supports without confusing.

Social Worker

Caseloads are heavy. Cortex remembers each family's history, agency policies, and resources you've already tried.

Designer

Design systems, critique notes, client preferences pile up. Cortex hands Claude the right reference at the right moment.

Marketer

Brand voice, campaign patterns, what worked last quarter. Cortex keeps Claude on-brand without the context-paste.

Biologist

Protocols, prior results, IACUC feedback. Cortex hands Claude the decisions you already made.

Engineer

Stack choices, conventions, past architectural calls. Cortex keeps Claude inside the rails you already built.

Lawyer

Firm policy, client precedent, redline patterns. Cortex surfaces your positions — not generic contract advice.

Accountant

GL rules, close-process quirks, client conventions. Cortex remembers the classifications so you don't re-explain each month.

Teacher

Lesson plans, accommodations, what worked last term. Cortex keeps Claude aligned with your curriculum, not a generic one.

Claude Code Developer

Terminal sessions, project rules, corrections you've made a dozen times. Cortex stops the groundhog day.

Free to run. Honest about what's next.

Local
Free
forever · the whole control panel on your Mac
  • Memories, agents, skills, projects & sessions
  • Automatic capture, decay & consolidation
  • Works in Claude Code, Claude Desktop & Cowork
  • AI features optional, on your own Claude subscription
  • Your data stays on your Mac
Download the Beta
Cloud Sync Coming
$49
/ year · planned
  • Zero-setup sync to claude.ai and mobile
  • Across all your devices, nothing to host
  • Everything in Local, plus cross-surface reach
Notify me →

Want cross-device sync today? Power users can self-host the Connector free on their own Cloudflare — about 15 minutes to set up. Ask us for the setup guide →

Not downloading yet? Stay in the loop.

Cortex is free to download today. Leave your email and we'll ping you when new features ship — or when the cloud version lands. No newsletter, just the note.

You're on the list. We'll send one note when Cloud Sync or 1.0 lands. Nothing else. In the meantime, Cortex is free to download today.

Your email is used only for these updates. See our privacy policy.

FAQ

Doesn't Claude already have memory?

It does — but per surface, and you maintain it. Claude Code keeps one memory, Claude Desktop another, your phone a third, and none of them follow you between machines. Cortex is the layer underneath that makes it one memory, kept in sync everywhere Claude runs.

Do I have to tell Cortex what to remember?

No. When a Claude Code session ends, Cortex saves the substantive moments — decisions, corrections, preferences — on its own. You can still browse, pin, or edit anything, but day to day it runs itself. Don't want automatic capture? One switch turns it off.

Does Cortex use AI? Does it cost anything?

The core — capturing, organizing, and surfacing your memory — runs on your Mac and is free, with no AI. A few optional features can call a language model for a nicer result; when you turn one on, it runs through your own Claude subscription using the Claude Code you already have. Cortex has no API key and never charges you for AI. If you don't have Claude Code, those features simply don't appear — everything else works the same.

Is it really free to download right now?

Yes. Cortex is in public beta and free to download for macOS. Expect a few rough edges — and updates that arrive automatically as we fix them.

Does my data leave my Mac?

No, not unless you choose it. By default Cortex makes no network calls and has no servers — your memory lives in a SQLite database and a readable markdown mirror under ~/.cortex/. The only exceptions are opt-in: the optional AI features (which run through your own Claude, never through us) and the self-host Connector you'd set up yourself.

How do updates work?

Cortex updates itself automatically and tells you when a new build is ready. You can also check manually in Settings → About, where the release notes live.

See all questions →