Supercode

AI-Powered SWE Agent

A monorepo containing the Supercode dashboard, documentation, terminal web client, CLI coding agent, and shared packages.

---

Star History

---

Overview

Supercode is a full-stack AI-powered development platform built as a Bun + Turborepo monorepo. It includes a web dashboard, an MDX documentation site, an AI coding agent CLI, a terminal-style web client, and a parallel project for fine-tuning open-source LLMs.

• Dashboard (apps/web) — Next.js dashboard for managing repositories, viewing analytics, and GitHub OAuth
• Documentation (apps/docs) — MDX-based documentation site
• Terminal Web Client (apps/supercode-cli/client) — Browser UI that mirrors the CLI experience
• Supercode CLI (apps/supercode-cli/server) — AI-powered coding agent published to npm as supercode
• API Server (apps/api) — Scaffolding for a shared backend
• Open Model (supercode-openmodel/) — Fine-tuning open-source LLMs (Qwen3, GLM-4) for coding tasks

Architecture

This is a monorepo managed with Turborepo and Bun workspaces. Workspaces are defined in package.json as apps/*, apps/supercode-cli/*, and packages/*.

Applications

App	Description	Port
apps/web	Next.js 16 dashboard (supercli.com)	3000
apps/docs	Next.js MDX documentation site	3001
apps/supercode-cli/client	Terminal web client UI (Next.js)	3002
apps/supercode-cli/server	AI coding agent — also published as the supercode npm CLI	—
apps/api	Shared API server (scaffolded)	TBD

Packages

Package	Description
@super/db	Prisma client + schema for the dashboard database
@super/auth	Better-Auth configuration (server + client)
@super/ui	Shared UI components
@super/sdk	Internal SDK
@super/config	Shared configuration (ESLint, TS, etc.)
@super/dashboard	Dashboard-specific component library
@super/db-terminal	Prisma client + schema for the terminal CLI's separate database
@super/claude-sdk	Claude / Anthropic AI SDK wrapper
@super/embeddings-sdk	Embeddings provider SDK (Gemini, etc.)
@super/skills	Reusable AI agent skills shared across apps

Terminal Stack 🖥️

The apps/supercode-cli workspace contains two sub-apps that together form the Supercode terminal product:

apps/supercode-cli/
├── client/   # Next.js 16 web client — terminal.supercli.com
└── server/   # Bun-powered AI coding agent, exposed as the `supercode` CLI

• Client is a Next.js app that mirrors the terminal experience in the browser, intended to be deployed to terminal.supercli.com.
• Server is a real installable CLI (supercode on npm) built with Bun and the AI SDK. It supports multiple model providers (OpenRouter, Anthropic, Google) and ships with a tool system (file read/write, command execution, search, etc.).

Run the CLI locally with:

bun run supercode              # dev (from root)
bun run dev:terminal           # run the web client
bun run dev:terminal-server    # run the CLI dev loop

Supercode Open Model 🧠

We're building a coding-focused LLM by fine-tuning open-source models for the Supercode CLI. This project runs in parallel with the main platform.

Goal

Fine-tune open models to create an AI assistant specialized in:

• Code generation and completion
• Debugging and error fixing
• Code explanation and documentation
• Multi-language programming support

Training Strategy

We're using a dual-track approach to train two models simultaneously:

Track	Infrastructure	Model	Method	Estimated Cost
Track 1	Tinker (Managed)	Qwen3-8B	LoRA	$150 free credits
Track 2	Modal + Axolotl	GLM-4-9B	LoRA/QLoRA	~$15-20

Project Structure

supercode-openmodel/
├── training/
│   ├── data/                    # Dataset preparation scripts
│   ├── tinker_qwen3/            # Tinker + Qwen3 training code
│   └── evaluation/              # Model evaluation scripts
├── models/                      # Model checkpoints
├── config/                      # Training configurations
└── pyrightconfig.json

Training Datasets

We use high-quality coding datasets:

Dataset	Size	Purpose
CodeAlpaca	20K	Instruction tuning
OpenCodeReasoning	100K+	Reasoning + coding
CodeFeedback	156K	Instruction-response

Quick Start

1. Set up Tinker (Track 1)

   pip install tinker
   export TINKER_API_KEY="your-key"
   cd supercode-openmodel/training/tinker_qwen3
   python train.py

2. Set up Modal (Track 2)

   pip install modal
   modal token new
   # Configure HuggingFace token in Modal secrets
   modal run --detach src.train --config=config/glm4.yml

Current Status

🚧 In Progress — Actively training and evaluating models. The best-performing model will be deployed to the Supercode CLI.

---

Tech Stack

Framework & Runtime

• Next.js 16 — React framework with App Router
• React 19 — UI library
• TypeScript 5 — Type safety
• Bun 1.2+ — JavaScript runtime & package manager

Monorepo

• Turborepo 2 — Build system for monorepos

Database & ORM

• PostgreSQL — Primary database
• Prisma 7 — Database ORM (separate schemas for db and db-terminal)
• Pinecone — Vector database

Authentication

• Better Auth — Authentication system
• GitHub OAuth — Social login

Background Jobs

• Inngest — Event-driven background jobs in apps/web

UI & Styling

• Tailwind CSS v4 — Utility-first CSS
• Radix UI + Base UI — Component primitives
• Framer Motion — Animations
• Lucide Icons — Icon library
• shadcn/ui — Component scaffolding for supercode-cli/client

State & Data

• TanStack Query — Server state management
• React Hook Form — Form handling
• Zod — Schema validation

AI & ML

• AI SDK v6 — Unified AI provider interface
• OpenRouter — Multi-model routing
• Anthropic Claude — Default model
• Google Gemini — Embeddings
• Vercel Minimax AI Provider — Additional model provider

CLI

• Commander — CLI argument parsing
• Clack + Boxen + Chalk — Terminal UI primitives
• Marked + marked-terminal — Markdown rendering in terminal

Getting Started

Prerequisites

• Bun 1.2+ (Install)
• PostgreSQL database (Neon, Supabase, or local)
• Git

Installation

1. Clone the repository

   git clone https://github.com/yashdev9274/supercli.git
   cd supercli

2. Install dependencies (this also runs db:generate via postinstall)

   bun install

3. Set up environment variables

   cp .env.example .env

   Edit .env with your configuration. At minimum, the dashboard requires:

   DATABASE_URL="postgresql://..."
   BETTER_AUTH_SECRET="your-secret-key"
   BETTER_AUTH_URL="http://localhost:3000"
   GITHUB_CLIENT_ID="your-github-oauth-id"
   GITHUB_CLIENT_SECRET="your-github-oauth-secret"

4. Set up the database

   cd packages/db
   bun run db:generate
   bunx prisma migrate dev
   cd ../..

5. Start development servers

   # Start everything
   bun run dev
   
   # Or start specific apps
   bun run dev:web              # Dashboard on http://localhost:3000
   bun run dev:docs             # Docs on http://localhost:3001
   bun run dev:terminal         # Terminal web client
   bun run dev:terminal-server  # CLI agent dev loop

bun run dev:api # API server

6. **Open your browser**
- Dashboard: [http://localhost:3000](http://localhost:3000)
- Documentation: [http://localhost:3001](http://localhost:3001)

### GitHub OAuth Setup

1. Go to [GitHub Developer Settings](https://github.com/settings/developers)
2. Create a new OAuth App
3. Set Authorization callback URL: `http://localhost:3000/api/auth/callback/github` (for local development). In production, use your deployed domain, e.g., `https://your-app.com/api/auth/callback/github`.
4. Copy Client ID and Client Secret to your `.env` file

## Project Structure

supercli/ ├── apps/ │ ├── web/ # Dashboard → supercli.com │ │ ├── app/ # Next.js App Router pages │ │ ├── components/ # React components │ │ ├── hooks/ # Custom React hooks │ │ ├── lib/ # Utility libraries │ │ ├── modules/ # Feature modules │ │ ├── inngest/ # Background job functions │ │ └── public/ # Static assets (og-image, etc.) │ │ │ ├── docs/ # MDX documentation site │ │ ├── app/ # App router pages │ │ ├── components/ │ │ ├── content/ # MDX documentation files │ │ └── lib/ │ │ │ ├── supercode-cli/ │ │ ├── client/ # Next.js terminal web UI → terminal.supercli.com │ │ │ ├── app/ │ │ │ ├── components/ │ │ │ └── lib/ │ │ └── server/ # Bun CLI agent (published as supercode on npm) │ │ ├── src/ │ │ │ ├── cli/ # CLI commands + chat loop │ │ │ ├── service/ # AI provider service layer │ │ │ ├── tools/ # Tool implementations │ │ │ ├── lib/ # Shared libraries │ │ │ ├── config/ # Config + env handling │ │ │ └── types/ │ │ └── prisma/ # Terminal DB schema (uses @super/db-terminal) │ │ │ ├── api/ # Shared API server (scaffolded) │ ├── packages/ │ ├── db/ # Prisma client for dashboard DB │ │ └── prisma/schema.prisma │ │ │ ├── db-terminal/ # Prisma client for terminal DB (separate schema) │ │ └── prisma/schema.prisma │ │ │ ├── auth/ # Better-Auth config │ │ └── src/ │ │ ├── server.ts │ │ └── client.ts │ │ │ ├── claude-sdk/ # Claude / Anthropic provider wrapper │ ├── embeddings-sdk/ # Embeddings provider wrapper │ ├── skills/ # Shared AI agent skills │ ├── ui/ # Shared UI components │ ├── sdk/ # Internal SDK │ ├── config/ # Shared ESLint/TS config │ └── dashboard/ # Dashboard-specific components │ ├── supercode-openmodel/ # LLM fine-tuning project │ ├── training/ │ │ ├── data/ │ │ ├── tinker_qwen3/ │ │ └── evaluation/ │ ├── models/ # Checkpoints │ └── config/ │ ├── scripts/ # Repo-level scripts ├── turbo.json # Turborepo config ├── package.json # Root package (workspaces, scripts) ├── bun.lock # Bun lockfile ├── .env.example # Environment template ├── CONTRIBUTING.md # Contribution guide └── README.md # This file

## Available Scripts

Run from the **repo root** unless otherwise noted.

### Development

| Script | What it does |
|--------|--------------|
| `bun run dev` | Start all dev servers (Turborepo) |
| `bun run dev:web` | Dashboard only (port 3000) |
| `bun run dev:docs` | Docs only (port 3001) |
| `bun run dev:terminal` | Terminal web client only |
| `bun run dev:terminal-server` | CLI agent dev loop |
| `bun run dev:api` | API server only |
| `bun run dev:video` | Remotion studio |
| `bun run supercode` | Run the published CLI in dev mode |
| `bun run supercode:prod` | Run the built CLI from `dist/` |

### Build & Quality

| Script | What it does |
|--------|--------------|
| `bun run build` | Build all apps and packages |
| `bun run lint` | Run ESLint across all packages |
| `bun run typecheck` | Run TypeScript checks |

### Database

From `packages/db/`:

```bash
bun run db:generate      # Clean and regenerate Prisma client
bun run db:migrate       # Deploy migrations to the configured DB
bunx prisma studio       # Open Prisma Studio
bunx prisma migrate dev --name migration_name  # Create a new migration

The terminal CLI has its own database and Prisma schema under packages/db-terminal/:

cd packages/db-terminal
bun run db:generate
bunx prisma migrate dev --name migration_name

Environment Variables

See .env.example for the full template. The variables below are documented for convenience.

Required (Dashboard)

Variable	Description
DATABASE_URL	PostgreSQL connection string
BETTER_AUTH_SECRET	Session encryption secret (generate with openssl rand -base64 32)
BETTER_AUTH_URL	Base URL of the dashboard (e.g., http://localhost:3000 for dev)
GITHUB_CLIENT_ID	GitHub OAuth app client ID
GITHUB_CLIENT_SECRET	GitHub OAuth app secret
NEXT_PUBLIC_BETTER_AUTH_URL	Public auth URL exposed to the client
NEXT_PUBLIC_APP_BASE_URL	Public base URL of the app (used for OAuth callbacks, absolute links)

Optional

Variable	Description
GOOGLE_GEMINI_API_KEY	Google Gemini (embeddings + text)
PINECONE_DB_API_KEY	Pinecone vector database
HUGGING_FACE_TOKEN	Hugging Face API token
TINKER_API_KEY	Tinker API key for LLM fine-tuning
VERCEL_OIDC_TOKEN	Auto-generated by Vercel CLI for deployment

Contributing

We welcome contributions! Please see the Contributing Guide for details on:

• Development setup
• Code style guidelines
• Commit conventions
• Pull request process

License

This project is licensed under the MIT License — see apps/supercode-cli/server/LICENSE for the published CLI's terms. Internal application code is currently private and proprietary to the Supercode team.

---

Built with ❤️ by the Supercode team!