Archon

Open source under Apache-2.0. Forks and pull requests are welcome.

Archon is a multi-agent application delivery platform that turns a prompt into a versioned execution with artifacts, live preview, evaluation, and governance surfaces. It is designed around traceability, recovery, scoring, and auditability rather than a single prompt-to-page interaction.

Live Demo • Video Walkthrough • Showcase Gallery

What Archon Demonstrates

• versioned artifact pipeline: brief, plan, code, preview, and factsheet per execution
• model-agnostic orchestration across Anthropic, Google, IBM, OpenAI, and local Ollama-backed workflows
• live version history with preview, restore, and prompt lineage
• runtime repair and build-recovery work for brittle generated React/TypeScript/Vite outputs
• automated eval loops with vision-based scoring and benchmark-driven iteration
• governance surface with model registry, quality scoring, and human-review gating

Demo

Walkthrough: prompt to generated app, version history, preview, and governed delivery flow.

Standout Enterprise Surface

The governance / factsheet screen is one of the strongest product surfaces in the repo because it immediately communicates auditability, traceability, and enterprise delivery posture.

What it shows:

• prompt quality scoring
• build confidence scoring
• model registry visibility across providers
• human-review gating
• a client-facing print/export surface

Selected Generated Examples

These examples represent the strongest generated outputs currently included in the repository.

Crypto Portfolio Dashboard

Prompt: "Build a crypto portfolio tracker with real-time prices, holdings table, and activity feed"

Halo Fan Page

Prompt: "Build a premium Halo fan page centered on Master Chief, Cortana, and the Arbiter. Include a cinematic hero, polished character dossiers, a legendary weapon showcase, and an explorable ringworld atlas."

SaaS Landing Page

Prompt: "Build a landing page for an AI-powered writing assistant with features, pricing, and testimonials"

More examples live in docs/SHOWCASE_GALLERY.md.

Core Product Surfaces

Versions And Live Preview

This is the most differentiated surface in the repo: execution lineage, preview refresh, prompt history, and restore behavior in one place.

Multi-Agent Pipeline

The system persists the intermediate work, not just the final output.

Artifact Viewer

Artifacts remain visible and versioned, which makes the pipeline easier to inspect, review, and restore.

Model-Agnostic Design

Archon is intentionally designed to route across providers rather than depend on a single model story.

• Anthropic Claude for premium code generation and showcase-quality runs
• Google Gemini / Vertex AI for planning, design direction, and image workflows
• IBM Watson NLU for governance, prompt analysis, and audit framing
• OpenAI in adjacent or legacy agent paths
• Local Ollama for lower-cost repeated evaluation and prompt-improvement loops

This design supports:

• provider abstraction
• premium vs economical routing
• cloud and local evaluation modes
• stable artifact/governance layers even as model choices change

Operating Modes

The repo separates cheap repeated iteration from premium final demos:

• bulk profile for repeated eval and reliability work
• showcase profile for a small number of premium hero builds

# Bulk reliability pass
python eval/eval_loop.py --config eval_config.json --profile bulk --archetype dashboard --runs 5 --skip-image-gen

# Premium showcase pass
$env:ENGINEER_MODEL = "claude"
$env:ENGINEER_CLAUDE_MODEL = "claude-opus-4-6"
$env:DESIGN_IMAGE_MODEL = "imagen-4.0-ultra-generate-001"
python eval/eval_loop.py --config eval_config.json --profile showcase --archetype game --runs 1

Architecture

User Prompt
  -> NLU / prompt analysis
  -> requirements artifact
  -> plan artifact
  -> design + image workflow
  -> code generation
  -> build / preview
  -> eval scoring
  -> governance factsheet
  -> version timeline + restorable execution

The result is a system where each run has visible lineage instead of a single opaque output.

Quick Start

Prerequisites

• Python 3.11+
• Node.js 20.9+ (Node.js 22 recommended)
• npm 10+

Install

git clone https://github.com/aiedwardyi/archon
cd archon

# Python backend
python -m venv venv
source venv/bin/activate            # macOS/Linux
# Windows PowerShell: .\venv\Scripts\Activate.ps1
pip install -r backend/requirements.txt

# Frontends (each has its own node_modules)
npm --prefix frontend ci
npm --prefix frontend-studio ci
npm --prefix frontend-consumer ci

Configure

Copy the example environment file:

cp backend/.env.example backend/.env
# Windows PowerShell: Copy-Item backend/.env.example backend/.env

The example defaults to OFFLINE_MODE=true. This provider-free path creates a PRD, plan, buildable React workspace, and live preview without API credentials. It suppresses configured provider calls, including Watson analysis. The generated workspace installs its own npm dependencies during the first build.

For live generation, set OFFLINE_MODE=false and configure Gemini through either GENAI_API_KEY or Vertex AI with VERTEX_AI_PROJECT and VERTEX_AI_LOCATION. Gemini or Vertex is required for requirements and planning. The Engineer Agent also defaults to Gemini.

To use Claude for engineering, add ANTHROPIC_API_KEY and set ENGINEER_MODEL=claude. Google OAuth and IBM Watson integrations remain optional.

Run

Start the backend, then run one or more UI surfaces. Each service needs its own terminal.

• Studio on port 3000: developer workspace and pipeline artifacts
• Consumer on port 3002: prompt-first application builder
• Enterprise on port 8080: governance, factsheets, and audit history

# Terminal 1: backend (http://localhost:5000)
source venv/bin/activate            # macOS/Linux
# Windows PowerShell: .\venv\Scripts\Activate.ps1
python backend/app.py

# Terminal 2: Studio UI (http://localhost:3000)
npm --prefix frontend-studio run dev

# Terminal 3: Consumer UI (http://localhost:3002)
npm --prefix frontend-consumer run dev

# Terminal 4: Enterprise UI (http://localhost:8080)
npm --prefix frontend run dev

Verify

python -m pytest -q
npm --prefix frontend run build
npm --prefix frontend-studio run build
npm --prefix frontend-consumer run build

Public Demo Deployment

For a safe public portfolio deployment, build the frontend in static demo mode instead of exposing the live builder:

cd frontend
$env:VITE_PUBLIC_DEMO_MODE = "true"
npm run build

The repo includes amplify.yml for a frontend-only Amplify deployment that serves the read-only showcase page and keeps backend/model execution paths private.

Further Reading

• Engineering overview
• Showcase gallery
• Technical reference
• Roadmap

License

Licensed under the Apache License, Version 2.0. See LICENSE.