reddypranith80-max/ghost-ai

Give it a goal and a tech stack. Watch a crew of specialised AI agents plan, build, test, commit and ship.
▶ Live demo ·
Dashboard ·
Demo video page ·
Download notebook ·
Book (PDF)
## Demo (90 seconds)
Your viewer cannot play inline video. Use the links below.
GitHub does not always play a repository .webm inline, so use whichever link works for you:
Watch on the live demo page:
Raw video file (about 11 MB, 89 s): public/demo/ghost-ai-demo.webm
Re-record it yourself: npm run demo:record (script: scripts/record-demo.mjs)
What you will see in 90 seconds
1. The marketing site: hero, how it works, the agent crew, pricing and docs pages.
2. The dashboard with sample projects, stats and a live activity feed.
3. The New project wizard: name, goal, stack picker, create.
4. Run next step moving tasks across the board (todo, running, review, done) and creating commits.
5. The agent roster, the git-log timeline and the flow graph of the 8-agent pipeline.
6. Exports: the file tree and a ZIP download of the whole project.
## Table of contents
What is Ghost AI
How it works
The crew
Every major component explained
Architecture
Tech stack
Project structure
Run it
Run the tests
Troubleshooting
Fork and clone
Download the notebook
Book
Honest status and known limitations
Contributing
Credits
## What is Ghost AI
The problem. Going from "I have an app idea" to a planned, structured project means juggling architecture, backend, UI, deployment, testing, security, docs and review. It is easy to lose track of who is doing what and what is done.
The idea. Ghost AI models that work as a crew of eight specialised lead agents, each with three side agents. You describe an app and pick a stack. Ghost AI creates a project with a task plan, then walks it through a pipeline (Architect, Backend, Frontend, DevOps, QA, Security, Docs, Reviewer). Every step moves a task forward and records a commit in a git-style log. You can watch it on a kanban board and a flow graph, and export everything.
Who it is for. Learners and builders who want to see what a multi-agent build pipeline looks like, people who want a polished full-stack Next.js reference project (API, database, auth, CI/CD, docs, tests), and anyone evaluating the concept.
> Important and honest: the agents are simulated. No real AI model is called. Tasks, progress and commits are generated by deterministic code in src/lib/sim.ts and src/lib/repo.ts. Ghost AI shows the workflow and the product shape; it does not write your app for you.
## How it works
mermaid
flowchart LR
A["You: goal + tech stack"] --> B["Create project: task plan per module"]
B --> C["Run next step"]
C --> D["Architect"] --> E["Backend"] --> F["Frontend"] --> G["DevOps"]
G --> H["QA"] --> I["Security"] --> J["Docs"] --> K["Reviewer"]
K --> L["Output: tasks, commit log, export ZIP or folder"]
Each click on Run next step advances the pipeline: a task in review becomes done (and a commit is logged), a running task moves to review, and the next todo task starts running. When every task is done the API answers 409 pipelinecomplete.
> Explain like I am 5. Imagine eight helpers in a toy factory. One draws the plan, one builds the engine, one paints it, one packs the box, one tests every button, one checks for dangers, one writes the instructions, and the last one says "OK to ship". You give the factory an idea, and you watch each helper take their turn. Longer version with analogies: docs/EXPLAIN-LIKE-IM-5.md.