Skip to main content

​
What is CogOS?

CogOS is a structured, persistent memory system for LLM-powered agents. It enables agents to truly remember — without flooding the context window. Through an agentic, schema-driven architecture, the Cognitive Memory (CM) agent uses frontier models to capture high-dimensional abstract signals about users that naive retrieval approaches miss entirely.

Quick Start

Get CogOS up and running in minutes.

Architecture

Understand how the system works end-to-end.

Python API

Use CogOS in your Python applications.

REST API

Integrate via the built-in HTTP server.

​
Key Results

  • 50% on PersonaMem v2 MCQ benchmark — state-of-the-art among persistent memory systems
  • 70% reduction in conversation token consumption on internal evaluation, with full contextual fidelity preserved
  • Plug-and-play context reconstruction — drop a user’s schema into any new session and the model recovers a rich portrait of the user from the very first message

​
Core Idea: Schema Call Protocol (SCP)

CogOS treats agent memory as a self-describing schema — like a file system for cognition: 
user_profile.identity.name        → "Alice"    (User's full name)
user_profile.preferences.hobbies  → "Hiking"   (Hobbies and interests)
user_profile.work.role            → "Engineer"  (Job title)
Agents interact with memory through 4 standard operations: recall_schema, update_schema, create_schema_field, and register_schema.

Learn more about SCP

Dive into the Schema Call Protocol architecture.

​
Vision: Cognition, Not Just Memory

Memory is a means, not the goal. CogOS aims to build a complete cognitive map of the user — one that captures not only what they explicitly say, but who they fundamentally are. The next frontier is hidden state modeling — moving beyond what users say or do, toward understanding what they mean and feel. This is the path to AI companions and assistants with genuine emotional intelligence.