AI-orchestrated software development platform.
You write a spec. Pantheon assembles a team of specialist AI agents — Controller, Architect, Coders, Reviewer, Auditor, Banker — and runs them against your project in coordinated sprints. Agents collaborate, spawn sub-teams when needed, negotiate conflicts, and get reviewed at every milestone. You watch it happen in real time.
v0.9.0-beta — This is an early public release. Core execution works. Some advanced features (background jobs, Git integration, live preview) are on the roadmap.
You write a spec
↓
Controller Agent scores it 1–5 and assigns a resource tier
↓
Controller plans sprints, tasks, and assembles the right team
↓
Agent team executes — Coders build, Reviewer inspects, Architect designs
↓
Banker Agent monitors every token and dollar in real time
↓
Conflict detected → Meeting Protocol → Auditor approval
↓
Sprint complete → Auditor reviews → gate opens or corrections required
↓
Deliverable files extracted into the Files tab → download as zip
| Tier | Spec Score | Agents | Sprints | Default Budget |
|---|---|---|---|---|
| micro | 1/5 | 3 | 1 | $1 |
| small | 2/5 | 8 | 3 | $5 |
| medium | 3/5 | 20 | 8 | $25 |
| large | 4/5 | 50 | 20 | $100 |
| enterprise | 5/5 | ∞ | ∞ | $250 |
Better spec → more powerful team. Write with precision.
- Node.js 18.17 or later
- pnpm (
npm install -g pnpm) - Supabase account — free tier at supabase.com
- At minimum one LLM provider API key:
- Google AI Studio (Gemini — free tier, used by Controller/Auditor)
- Fireworks AI (Llama models — free tier, used by Coders/Banker)
- Anthropic (Claude — optional, not default in v0.9)
git clone https://github.com/iAmBalanceAR/pantheon.git
cd pantheonpnpm install- Go to supabase.com and create a new project
- Wait for it to provision (about 1 minute)
- Go to Project Settings → API and copy:
- Project URL →
NEXT_PUBLIC_SUPABASE_URL - anon / public key →
NEXT_PUBLIC_SUPABASE_ANON_KEY - service_role / secret key →
SUPABASE_SERVICE_ROLE_KEY
- Project URL →
In your Supabase project, go to SQL Editor → New query.
Copy the entire contents of supabase/schema.sql and run it.
This creates all tables, RLS policies, indexes, and realtime subscriptions in a single pass. It is idempotent — safe to run again if anything goes wrong.
cp .env.example .env.localOpen .env.local and fill in your values:
# Required — from your Supabase project settings
NEXT_PUBLIC_SUPABASE_URL=https://your-project-ref.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key
# Required — at least one LLM provider
GOOGLE_API_KEY=AIza... # for Controller + Auditor agents
FIREWORKS_API_KEY=fw_... # for Coder + Banker agentsSee .env.example for the full list of options including model overrides.
In your Supabase dashboard:
- Go to Authentication → Providers
- Ensure Email provider is enabled (it is by default)
- Under Authentication → URL Configuration, set:
- Site URL:
http://localhost:3000 - Redirect URLs:
http://localhost:3000/**
- Site URL:
pnpm devOpen http://localhost:3000.
Create an account, write a spec, and run your first project.
The Controller scores your spec 1–5 before assembling the team. Score directly determines team size and sprint count. Here is how to hit each tier:
Score 1 (micro): A single vague sentence with no stack or constraints.
"Build me a todo app."
Score 3 (medium): Clear goal, implied stack, some acceptance criteria.
"Build a React landing page for a SaaS product. Include a hero section, feature grid, pricing table with 3 tiers, and a contact form that posts to /api/contact. Use Tailwind CSS. Mobile-first layout."
Score 5 (enterprise): Formal spec with architecture, acceptance criteria, performance requirements, security considerations, and explicit stack.
Tips for better specs:
- Name your stack explicitly:
"Next.js 14, Supabase, Tailwind CSS"beats"a web app" - State acceptance criteria:
"Clicking Submit posts to /api/contact and shows a success toast" - Mention non-functional requirements:
"Must work without JavaScript","mobile-first layout" - Flag scope boundaries:
"Do not add authentication","this sprint only covers the landing page"
| Role | Responsibility |
|---|---|
| Controller | Scores the spec, plans sprints, assembles the team, orchestrates execution |
| Coder | Implements tasks — writes code, markup, copy, data models |
| Architect | High-level design, system structure, technical decisions (medium+ tiers) |
| Reviewer | Code review, standards enforcement, output quality (medium+ tiers) |
| Auditor | Sprint review and gate — approves or rejects before advancing |
| Banker | Monitors token and dollar spend, enforces budget guardrails |
| Researcher | Gathers context, investigates unknowns, informs other agents |
| Mediator | Resolves inter-agent conflicts via the meeting protocol |
| Custom | User-defined agents — assign any role from the Skills Library |
The Agents page includes a built-in Skills Library with 14 specialist agent types across four categories:
- Development: Mobile Developer, DevOps Engineer, Security Engineer, Data Engineer, API Designer, Python Developer, Database Administrator
- Creative: Copywriter, Technical Writer, UX Writer
- Analysis: Data Analyst, Business Analyst
- Process: QA Engineer, Product Manager
Install a skill to make it available as a named agent type in your projects. Installed skills can be edited to inject your own domain context.
Agents emit deliverable files inside <file path="..."> blocks in their task output. Pantheon extracts these automatically. Browse and download them as a zip from the Files tab on any project page.
For web projects:
- Agents output
index.html,styles.css,scripts.jsas separate<file>blocks - Images use hotlinked URLs from free stock sources (Unsplash, Pexels, Shopify Burst) — no binary assets required
src/
app/
(auth)/
login/ Auth pages
signup/
(dashboard)/
projects/ Project list and new project wizard
projects/[id]/ Project overview with real-time updates
projects/[id]/chat Live agent communication feed
projects/[id]/files Deliverable files browser and zip download
projects/[id]/team Team visualization
projects/[id]/budget Banker dashboard
agents/ Skills library and agent roster customization
settings/ Platform configuration
api/
projects/ Project CRUD + spec analysis
projects/[id]/run Sprint execution (chains all sprints server-side)
agents/library/ Skills library browser API
agents/skills/ User installed skills API
settings/ Agent profiles and platform settings
lib/
supabase/ Client, server, and service role clients
llm/ Anthropic, Fireworks, Gemini provider wrappers
agents/
skills/ Core agent skill markdown files (immutable)
skills-library/ Installable specialist skills (14 included)
controller.ts Spec analysis and team planning
engine/
executor.ts Task execution loop
persist-task-files.ts Deliverable extraction
deliverable-refs.ts Multi-file consistency checker
types/index.ts Shared types and tier configuration
supabase/
schema.sql Complete database schema (run once to set up)
migrations/ Individual migration files (historical reference)
docs/ Architecture and development documentation
| Variable | Required | Description |
|---|---|---|
NEXT_PUBLIC_SUPABASE_URL |
✅ | Your Supabase project URL |
NEXT_PUBLIC_SUPABASE_ANON_KEY |
✅ | Supabase anon/public key |
SUPABASE_SERVICE_ROLE_KEY |
✅ | Supabase service role key (server-side only) |
GOOGLE_API_KEY |
✅* | Gemini API key — required for Controller/Auditor |
FIREWORKS_API_KEY |
✅* | Fireworks API key — required for Coder/Banker |
ANTHROPIC_API_KEY |
Optional | Claude API key (not used by default in v0.9) |
CONTROLLER_PROVIDER |
Optional | Override Controller provider (gemini/fireworks) |
CONTROLLER_MODEL |
Optional | Override Controller model |
AUDITOR_PROVIDER |
Optional | Override Auditor provider |
AUDITOR_MODEL |
Optional | Override Auditor model |
CODER_PROVIDER |
Optional | Override default Coder provider |
CODER_MODEL |
Optional | Override default Coder model |
BANKER_PROVIDER |
Optional | Override Banker provider |
BANKER_MODEL |
Optional | Override Banker model |
NEXT_PUBLIC_APP_URL |
Optional | App base URL (default: http://localhost:3000) |
*You need at least one of Gemini or Fireworks, but both are recommended for the default agent configuration.
| Layer | Technology |
|---|---|
| Framework | Next.js 14 (App Router) |
| Auth + Database | Supabase (PostgreSQL + Auth + Realtime) |
| Styling | Tailwind CSS + Radix UI |
| LLM Providers | Google Gemini, Fireworks AI, Anthropic |
| Language | TypeScript |
| Package manager | pnpm |
- Push the repository to GitHub
- Import the project in Vercel
- Add all environment variables from
.env.exampleto the Vercel project settings - In your Supabase project Authentication → URL Configuration, add your Vercel deployment URL to Redirect URLs
- Background job runner (async agent execution — non-blocking API routes)
- Live preview for web deliverables
- Git integration (agents commit to branches, merge conflict detection)
- Model selection UI per project (override which model each role uses)
- Webhook notifications (Slack, Discord on sprint completion)
- Team templates (save and reuse proven team configurations)
Contributions are welcome. Please open an issue before submitting a large pull request so we can discuss the approach.
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature - Make your changes and ensure
pnpm buildpasses - Open a pull request with a clear description of what changed and why
MIT — see LICENSE for details.