R H E A
Multi-model consensus with verifiable proofs

Try it · iOS Beta · v1.0.0 · EU Mirror

---

You submit a claim. Multiple AI models evaluate it independently. Rhea computes agreement scores, confidence intervals, and stores verifiable proofs. No single model gets the final word.

"Lipinski's Rule of Five predicts oral bioavailability"
         │
         ▼
┌─────────────────────────────────────────────┐
│  Tribunal (multi-model consensus engine)     │
│  ├─ Gemini 2.5 Flash  → agrees (0.82)      │
│  ├─ GPT-5.3           → agrees (0.76)      │
│  ├─ Claude Opus 4.6   → partially (0.54)   │
│  └─ Sceptic (red team) → challenge filed    │
├─────────────────────────────────────────────┤
│  Agreement: 71%  │  Confidence: 68%         │
│  Proof ID: ale_1709...  │  Stored in DB     │
└─────────────────────────────────────────────┘

Install

Server (one command)

curl -sL https://raw.githubusercontent.com/serg-alexv/rhea-project/main/deploy/setup.sh | bash

Works on macOS, Ubuntu, Debian, Alpine, Amazon Linux, Arch. Auto-detects your OS, installs Python 3.10+ if needed, creates a virtualenv, generates .env, starts the server on :8400.

Server (manual)

git clone https://github.com/serg-alexv/rhea-project.git
cd rhea-project
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Edit .env — add at minimum GEMINI_API_KEY (free: https://aistudio.google.com)
python3 src/tribunal_api.py
# http://localhost:8400

Windows

git clone https://github.com/serg-alexv/rhea-project.git
cd rhea-project
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
copy .env.example .env
# Edit .env — add GEMINI_API_KEY
python src\tribunal_api.py

Memory package (standalone)

pip install rhea-memory

Zero-dependency persistent memory for AI agents. Package docs →

Use it

Submit a claim for consensus

curl -X POST http://localhost:8400/tribunal \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_KEY" \
  -d '{"prompt": "Aspirin inhibits COX-2 selectively"}'

Returns agreement score, confidence interval, individual model votes, and a proof ID.

Red-team a claim

curl -X POST http://localhost:8400/sceptic \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_KEY" \
  -d '{"prompt": "CRISPR has no off-target effects"}'

The sceptic actively tries to disprove the claim.

Store a proof

curl -X POST http://localhost:8400/aletheia/submit \
  -H "Content-Type: application/json" \
  -d '{"claim": "...", "evidence": "...", "source": "tribunal"}'

Every consensus result gets a hash-chained proof you can verify later.

Create an account

curl -X POST http://localhost:8400/auth/signup \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com", "password": "your-password"}'

Returns a JWT token for authenticated API access.

What's in the box

Component	What it does
Tribunal API	Python server — multi-model consensus, red-team sceptic, proof store, JWT auth
rhea-memory	pip install rhea-memory — persistent memory for AI agents (MIT, zero deps)
iOS app	SwiftUI — 9 tabs: tribunal, radio, governor, tasks, atlas, pulse, bio, relay, settings
Command Centre	macOS — 3-column ops centre with menu bar widget
Atlas	Next.js — 3D web dashboard
Model bridge	10 providers, 42 models, 4 cost tiers (cheap → research)

API at a glance

Method	Path	What it does
POST	/tribunal	Multi-model consensus
POST	/sceptic	Red-team challenge
POST	/dialog	Interactive dialog
POST	/auth/signup	Create account
POST	/auth/login	Get JWT token
GET	/health	Server status
GET	/governor	Token budget
GET	/agents/status	Agent roster
GET	/feed/stream	SSE real-time feed
POST	/aletheia/submit	Submit proof
GET	/aletheia/search	Search proofs

Full API: 79 route handlers. See docs/ for architecture, decisions, and guides.

Architecture decisions

16 ADRs document every major choice. Here are the ones that matter most:

#	Decision	Why
002	Multi-model bridge over single provider	No vendor lock-in, consensus requires diversity
003	ADHD-optimized design	Real users have real constraints — UI must respect attention
008	Cheap-first model routing	Default to $0.001/query, escalate only when needed
009	Cost-aware agents	Every agent tracks its own spend
010	Memory budget + discomfort metric	Agents self-improve when context gets expensive

Full list: docs/decisions.md

Project structure

rhea-project/
├── src/                    # Python backend (tribunal, bridge, governor, auth)
├── packages/
│   ├── RheaKit/            # Swift framework (iOS + macOS views)
│   └── rhea-memory/        # pip-installable memory layer (MIT)
├── ios/                    # Xcode project, extensions (tunnel, keyboard)
├── rhea-atlas/             # Next.js dashboard
├── tools/                  # Rust CLI
├── deploy/                 # setup.sh, Docker Compose, infra
├── docs/                   # ADRs, architecture, guides
└── scripts/                # Automation

Contributing

git clone https://github.com/serg-alexv/rhea-project.git
cd rhea-project
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python3 src/tribunal_api.py
# Server on :8400 — make changes, test, submit PR

License

MIT. See LICENSE.

All third-party dependencies carry compatible licenses (MIT, BSD-3-Clause, Apache-2.0). See docs/licenses.md for the full dependency table.

---

_{"Logic is Fluid. Evidence is Durable. The Manifold is Smooth."
∇ > 0 ∨ ⊥}

_{github.com/serg-alexv/rhea-project
genesis · decisions · protocol}

_{TimeLabs NPO}