CortexClaw v4.5

CortexClaw

A memory system for AI agents, modeled on the brain. It keeps what matters across sessions and hands the agent only what each question needs - instead of re-reading everything, every time.

Read the full architecture → Want the deep version - every layer, the neuroscience, the internals? It all lives there.
12
Neural Layers
764
Memory Chunks
14,356
Synapses
92%
Token Efficiency

What It Is, In Plain Terms

What It Is

CortexClaw replaces flat memory files with small, searchable memory chunks that are retrieved on demand. The agent asks a question; CortexClaw returns just the relevant pieces.

Think of it as a personal search engine for an AI's memory.

How It Works

It sits alongside the model as external memory. When the agent needs to recall something, CortexClaw scores every chunk by meaning, recency, past usefulness, and how ideas connect - then returns the best matches.

Afterward, feedback flows back: what helped, what was noise. Over time it learns to retrieve better.

The Problem & The Fix

The Problem

AI agents forget everything between sessions. Loading whole memory files burns tokens and treats every note as equally important - so the useful patterns get lost in the noise.

The Fix

Model memory like the brain. Recent things stay sharp, important things harden into permanence, related memories link up, and idle time is spent consolidating. You load less and remember more.

What It Does

Remembers
Knowledge is stored as searchable chunks and pulled back by meaning, not by reloading files.
Forgets Well
Memories fade at different speeds - trivia decays fast, important patterns stay. Nothing useful gets buried.
Connects
Ideas that come up together wire together, so recalling one surfaces the others.
Consolidates
A background daemon replays and tidies memory during idle time, the way sleep does.
Stays Private
Scope routing keeps each workspace's memory isolated by default, with opt-in sharing.
Runs Lean
Pure Python, local embeddings, zero API cost - and a compressed index ~26× smaller than the raw vectors.