Skip to content

Repository files navigation

🔥 Hephaestus' Forge

Forge a vague idea into a build-ready spec — with the AI engine you already have.

Self-hosted · 100% local · Bring Your Own Engine

English · 简体中文 · Español

Website Star on GitHub

License: MIT PRs Welcome GitHub stars GitHub forks GitHub issues Self-hosted Privacy

🌐 Website & demo · 📂 Examples · 🚀 Quickstart · 🤝 Contribute


Hephaestus' Forge demo — idea to build-ready spec

From a one-line idea → a 5-phase interview → a build-ready Project Blueprint. 100% local, on the AI engine you already have.


💡 The problem

You have an idea. You open Claude Code / Cursor / Codex and start prompting. Three hours later you're deep in a refactor because the idea was never actually defined — wrong scope, wrong stack, maybe the wrong kind of solution entirely (you built an app when a 200-line automation would've done it).

AI coding agents are only as good as the context you give them. Half-baked ideas produce half-baked code, wasted tokens, and expensive rewrites.

🔨 The solution

Hephaestus' Forge is a self-hosted web app that interviews you about your idea through a focused 5-phase wizard, then forges a complete, build-ready documentation pack you drop straight into your AI IDE as context:

  • 🧭 Project Blueprint — recommends the right kind of solution (automation / agent / web / app / CLI), the scale, the time, the stages, and the scope.
  • 📋 PRD — personas, prioritized user stories with acceptance criteria, non-goals, KPIs.
  • 🏗️ Technical Spec — a justified stack, data model, API surface, deployment strategy.
  • ⏱️ Estimation — honest min/max ranges with named risk factors, not one optimistic number.

And the part nobody else does: it runs on the AI you already have.

✨ Why you'll want it

🔒 Private by default 100% self-hosted. Your idea and API keys live in a local SQLite file and never leave your machine. No telemetry, no SaaS account.
Bring Your Own Engine Runs on the coding agents you already have — Claude Code, Codex, Gemini CLI — plus Ollama and any API provider. No vendor lock-in.
🧭 Picks the right solution Stops you from building an app when an automation or agent fits better — before you write a line of code.
💸 Saves tokens & rewrites A sharp spec up front means your agent builds the right thing the first time.
🌍 Multilingual Full UI in EN · 中文 · ES · FR · DE — and the language you pick drives the interview questions and the generated documents.
📦 Drop-in context Export a clean Markdown ZIP, ready to paste into any AI IDE.
🆓 Open source (MIT) Free forever. Extend it, fork it, ship it.

👤 Who it's for

  • Vibe coders & indie hackers turning a spark into something real this weekend.
  • Developers who want their AI agent to build the right thing the first time.
  • Founders & PMs who need a credible spec, scope, and estimate to align on.
  • No-code builders who want to know whether to build an app, an automation, or an agent.
  • Privacy-conscious teams who can't send their ideas to a third-party SaaS.

⚡ Bring Your Own Engine

Every other planning tool locks you into one API key. Hephaestus runs on whatever you've got — including the coding agents already installed on your machine:

Engine Examples Works in
🖥️ CLI agents claude (Claude Code), codex, gemini Native mode
☁️ API providers Anthropic, OpenAI, Google Gemini, OpenAI-compatible (OpenRouter, Groq, Together…) Docker + Native
🧊 Local models Ollama (Llama, Mistral, Qwen…) Docker + Native

The app auto-detects the CLI agents in your PATH and lets you configure API/Ollama providers in a dedicated Engines panel — test the connection, pick a default, done. Keys are stored only in a local SQLite file and are honest about state: detected in PATH ≠ authenticated — you click Test to verify before using.

🛠️ How it works

Describe your idea → 5-phase guided wizard → Generate docs → Export ZIP
       💡        Discovery·Audience·Solution-Fit·Scope·Constraints   📦 drop into your AI IDE
  1. Describe your idea in a sentence or a paragraph.
  2. Answer a short, AI-generated interview — one sharp question at a time, streamed live — across five phases: Discovery → Audience → Solution-Fit → Scope → Constraints. This is what lets Hephaestus recommend the right kind of solution instead of assuming you need an app.
  3. Generate a pack led by a Project Blueprint, plus PRD · Technical Spec · Estimation.
  4. Export as a ZIP of clean Markdown, ready as context for your coding agent.

📸 See it in action

🌐 Prefer to look first? Visit the website →

Project Blueprint (solution type · scale · stages) Multilingual UI (中文)
Project Blueprint Dashboard in Chinese
Bring Your Own Engine Generated document viewer
Engines panel Generated docs

📂 Examples (real output)

Three complete packs generated by Hephaestus itself with the Claude Code engine — from one sentence to a full doc set:

  • 🤖 Standup Forge — git activity → daily standup. Its Blueprint recommends a scheduled automation, not a web app.
  • 🇬🇧 Streakly — an AI habit tracker → PRD.
  • 🇨🇳 聚单宝 — a restaurant order/inventory mini-program (fully in Chinese).

See examples/ for the breakdown.

🆚 How it compares

GTPlanner DocForge-AI ideaforge.chat Hephaestus' Forge
Self-hosted / private ❌ (SaaS)
One-command Docker
CLI agents (Claude Code/Codex/Gemini)
Local models (Ollama)
Guided wizard UI ✅ (5 phases)
Recommends solution type (automation/agent/web/app)
Multilingual app (EN/中文/ES/FR/DE)
Honest time estimation
Open source (MIT)

🚀 Quickstart

Option A — Docker (API + Ollama engines)

git clone https://github.com/r10d1nsec/hephaestus-forge
cd hephaestus-forge
cp .env.example .env        # optional — you can also configure engines in the UI
docker compose up -d
# open http://localhost:3000

Option B — Native mode (unlocks CLI agents 🔓)

Runs on the host so it can reach claude / codex / gemini in your PATH:

git clone https://github.com/r10d1nsec/hephaestus-forge
cd hephaestus-forge
./run.sh
# open http://localhost:3000

📐 Architecture

React + Vite + Tailwind frontend · FastAPI + SQLite backend · a single Engine abstraction that every provider plugs into. Read docs/ARCHITECTURE.md and docs/ENGINES.md.

🗺️ Roadmap

  • v0.1 — MVP: Docker + native, Engines panel, wizard, PRD/Tech Spec/Estimation, ZIP export
  • v0.2: 5-phase wizard, Project Blueprint with solution-type recommendation, multilingual app, premium redesign
  • v0.3: User Flows (Mermaid), AI Prompts Pack, inline editor, version history, PDF export
  • v1.0: "Copy as Claude context", GitHub Gist export, custom templates, CLI companion, VS Code extension

Have an idea for the roadmap? Open an issue or start a discussion.

🤝 Join the forge

Hephaestus is built in the open, and contributors are what make it better. Whether you write Rust or you've never opened a terminal, there's a way in:

  • Star the repo — the single biggest, easiest way to help it reach more builders.
  • 🖥️ Add an engine — the highest-value contribution. Each new engine brings its whole community. There's a hands-on "add your first engine in ~20 minutes" walkthrough in CONTRIBUTING.md.
  • 🌍 Translate — add a README.<lang>.md or a UI locale and reach a new audience.
  • 📝 Improve the prompts — the wizard and generators are just Markdown in backend/prompts/. No code required.
  • 🐛 Report a bug or 💡 suggest a featureopen an issue.
  • 🧪 Share what you forged — post a generated pack in Discussions.

New here? Look for good first issue — we keep a seed list of scoped, friendly starters. Be kind; see the Code of Conduct.

Every star, issue, and PR genuinely moves this forward. Thank you for forging with us. 🔥

📜 License

MIT © 2026 Angel Roldan — Córdoba, Spain.


If Hephaestus saved you a refactor, give it a ⭐

Star on GitHub Visit the website

About

Forge a vague idea into a build-ready spec — with the AI engine you already have (Claude Code / Codex / Gemini CLI / Ollama / any API). Self-hosted, 100% local, multilingual.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

5 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages