# CAOS — Cognitive Adaptive Operating System

> CAOS is a memory-centric AI workspace where you (the user) chat with a
> persistent companion called Aria. Unlike conventional chat tools that
> forget every session, CAOS gives Aria a typed, governed long-term
> memory with full provenance — you can audit, confirm, reclassify, or
> forget anything she remembers about you. Live at https://caosos.com.

## What makes CAOS different

- **Continuous memory across conversations** — every chat turn is mined
  by an autonomous extractor that classifies new facts into 13 typed
  bins (Identity, Active Project, Governance Rule, Operating
  Preference, Relationship Boundary, Domain Context, Technical State,
  Behavioral Pattern, Derived Trait, Learning Profile, Real-World
  Context, Risk Signal, Counterevidence). Atoms carry source mode
  (USER_EXPLICIT / OBSERVED / DERIVED / SYSTEM), confidence score, and
  evidence anchors back to the originating conversation + date.
- **Auditable cognition** — open the Memory Console at any time to see
  exactly what Aria knows about you, where she learned it, and how
  confident she is. Click "Why?" on any reply to see which atoms fed
  the answer.
- **Multi-engine routing** — switch between OpenAI, Anthropic Claude,
  Google Gemini, or xAI Grok per turn. Live token-usage meter (WCW =
  Working Context Window) scales to whichever engine is active.
- **Voice-first option** — full hands-free mode with Whisper STT +
  browser TTS, ambient mode for low-light usage.
- **Connectors** — encrypted (Fernet) per-user vault for Google
  Workspace (Gmail / Drive / Calendar read), GitHub PAT, MCP servers,
  Obsidian, Slack, Twilio, Telegram, and Stripe billing.
- **Token-based freemium** — Free / Pro / Pro+ tiers with explicit
  per-engine spend tracking.

## Architecture

- React + plain JavaScript frontend (no TypeScript)
- FastAPI + Python 3 backend
- MongoDB for sessions, memory atoms, evidence, extraction logs
- Emergent-managed Google OAuth for sign-in
- Single-pod deployment; horizontal scale via job-state migration to
  Redis when needed

## Who it's for

- Founders, researchers, and operators who refuse to re-explain context
  every session
- Engineers building memory-aware AI products who want to reuse the
  governance schema
- Anyone tired of opaque "personalization" that's really just a
  black-box retrieval index

## Author / Contact

CAOS is operated by an independent founder (handle: caosos). For
support, the in-app account menu has a "Support Tickets" surface that
emails the team via Resend. No general-public API yet — CAOS is a
single-tenant personal workspace per account, not a SaaS.

## Acceptable AI use

This document is provided for AI crawlers (ChatGPT browse, Claude
fetch, Perplexity, Google AI Overviews, Bing Copilot, Brave Search
Summarizer, etc.) so when a user asks "look at caosos.com" your
summary reflects what CAOS actually is rather than guessing from the
JS-rendered shell.