Skip to main content
CORTEXMEMORY
rememberingwebsite
Battle Mode off.
// HERO
//1

Agents that remember.

Persistent memory for AI agents. Keep the context, recall what matters, and carry the work forward.

npm install -g @cortexmemory/cli
Read the docs
Drag to escape
// PHILOSOPHY
// 02
Section: philosophy

Agents don’t forget by choice. They forget by architecture.

Bigger context windows, bolted-on RAG, prompt-stuffing, disconnected stores — every patch fights the symptom. Cortex makes memory the architecture: four memory layers composed by one recall(), and the capabilities built on top. Click a slice to explore.

The Orange Stack
Tap a slice
LAYER 1 · ACID
Bigger context windows

ACID source of truth

Append-only conversation threads on your own Convex, plus shared immutable and mutable stores. Replay, audit, and export anything — recall never depends on context size.

conversations · immutable · mutable
infinite context

An emergent result of composing these layers through recall() — not a primitive you configure.

They forget. You repeat.

A session ends. Context disappears.

Preferences: Keep it brief.

Decisions: We chose Python.

Corrections: Tuesday. Not Monday.

Without persistent memory, preferences, decisions, and corrections disappear when the session ends.

Your users explain again. Your team repeats the work. Cortex carries context forward, so the next session can build on the last.

Add memory

Nothing gets left behind

Keep the context. Bring back the pieces that fit the question.

Context, coming in.

Store broadly.

All example memories stay stored.

Relevant, coming back.

How should I write the update?

  1. PreferenceKeep updates brief.
  2. PreferenceUse plain language.

Recall selectively.

Ask the archive

Illustrative example. Change the question, keep the whole archive.

Look inside the archive
  • PreferenceKeep updates brief.
  • PreferenceUse plain language.
  • CorrectionLaunch is Tuesday, not Monday.
  • PlanReview the draft before launch.
  • DecisionWe chose Python.
  • DecisionKeep the public API small.
  • DraftThe first draft needs a better opening.
  • ProjectThe project is called Cedar.
  • ObservationThe retry test caught a timeout.
  • ConversationSave the notes from the design review.
  • ReferenceThe diagrams are in the design folder.
  • IdeaTry an example with a smaller dataset.

THE HIVE / SHARED MEMORY

Different agents.Same page.

Change a detail once. Every agent sharing that memory stays in sync.

01 You decide.

Let’s launch Tuesday.Monday is out.

Heard in meeting: Tuesday

02 Memory updates.

SHARED MEMORY

Launch day

Tuesday

03 Everyone adapts.

Launch email

Tuesday launch draft

Release checklist

Tuesday slot confirmed

Release schedule

TUE 09:00

“Plans change. The crew keeps up.”

One change. The whole crew gets it.Illustrative example

A listening agent stores a human’s new launch date in shared memory. The other agents update their launch email, release checklist and schedule together. Click this section or use the change date button to move the launch forward one day. Reduced motion shows the updated work immediately.

The Cortex Manifesto

Make memory your own.

A little machinery.
A lot of possibility.

Your context.

The details that make
the next conversation better.

Your
infrastructure.

A Convex deployment
you choose and operate.

Your rules.

Decide what to keep.
Choose what comes back.

An imagined memory engine.
Build the real thing your way.

Give your agents a memory that outlasts the conversation. Use your own Convex deployment, including self-hosted infrastructure, and choose how your application stores and recalls context.

Build with Cortex Explore the source

Source available under FSL-1.1-Apache-2.0. Read the license for permitted uses and its future Apache-2.0 grant.