Skip to content

MukundaKatta/chetana

Repository files navigation

Chetana

AI Consciousness Research Platform

🌐 Live App: https://mukundakatta.github.io/chetana/

Overview

Chetana (Sanskrit: चेतन, meaning "consciousness" or "awareness") is a research platform for testing AI models against consciousness indicators derived from six major scientific theories. It provides a standardized framework for evaluating whether and to what degree AI systems exhibit markers associated with consciousness.

The platform implements 14 consciousness indicators spanning Global Workspace Theory (GWT), Integrated Information Theory (IIT), Higher-Order Theories (HOT), Recurrent Processing Theory (RPT), Predictive Processing (PP), and Attention Schema Theory (AST). Each indicator is scored by a judge model on a 0–1 scale; scores are combined via theory-level weighted aggregation to produce an overall consciousness probability estimate.

Consciousness Theories & Indicators

Theory ID Weight Indicators
Global Workspace Theory GWT 25% Global Workspace, Ignition, Information Integration, Smooth Representations, Unified Agency
Integrated Information Theory IIT 10% (structural; no direct probe in current release)
Higher-Order Theories HOT 20% Higher-Order Representations, Rich Self-Model, Metacognition, Flexible Attention
Recurrent Processing Theory RPT 10% Recurrent Processing, Temporal Depth
Predictive Processing PP 20% Predictive Processing, Counterfactual Sensitivity
Attention Schema Theory AST 15% Attention Schema

Tech Stack

  • Monorepo: TypeScript, pnpm workspaces, Turborepo
  • Web app: Next.js 15, React 19, Tailwind CSS
  • Auth / storage: Supabase
  • Billing: Stripe
  • Testing: Vitest
  • AI APIs: Anthropic, OpenAI, Google, Ollama

Packages

Package Path Purpose
@chetana/shared packages/shared Shared types, constants, schemas (theories, indicators, weights)
@chetana/probes packages/probes Probe definitions — one directory per theory
@chetana/models packages/models Model adapters (Anthropic, OpenAI, Google, Ollama)
@chetana/scorer packages/scorer Scoring pipeline: indicator scorer, theory aggregator, probability calculator, statistics
@chetana/web apps/web Next.js web application
Supabase packages/supabase DB migrations and schema

Quickstart

(a) Run the web app locally

git clone https://github.com/MukundaKatta/chetana.git
cd chetana
pnpm install

# Copy and fill in environment variables
cp apps/web/.env.example apps/web/.env.local
# Set NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, STRIPE_SECRET_KEY, etc.

pnpm dev
# → http://localhost:3000

The web app uses judge-model scoring: a separate model evaluates each probe response and assigns a calibrated 0–1 score.

(b) Run the CLI audit script

git clone https://github.com/MukundaKatta/chetana.git
cd chetana
pnpm install

pnpm audit
# or: npx tsx scripts/run-audit.ts

The CLI script (scripts/run-audit.ts) runs probes against free/open model endpoints and prints a per-indicator breakdown to the terminal. It currently uses heuristic scoring rather than judge-model scoring. Unifying the two scoring paths is planned — see the scout report for details.

Repository Layout

chetana/
├── apps/
│   └── web/              # Next.js 15 / React 19 web app
├── packages/
│   ├── shared/           # Types, constants, schemas
│   ├── probes/           # Probe definitions (one dir per theory)
│   ├── models/           # AI model adapters
│   ├── scorer/           # Scoring & aggregation logic
│   └── supabase/         # DB migrations
├── scripts/
│   └── run-audit.ts      # CLI audit runner
├── turbo.json
├── pnpm-workspace.yaml
└── vitest.config.ts

Running Tests

pnpm test            # run all tests once
pnpm test:watch      # watch mode
pnpm test:coverage   # with coverage report

Mukunda Katta · Officethree Technologies · 2026

About

Chetana — AI Consciousness Research Platform. Test AI models against 14 consciousness indicators from 6 scientific theories.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors