Spec-Driven AI Development for Developers

Design and deploy agentic applications.

A pipeline of specialized LLM agents takes you from a rough idea — or an existing codebase — to full specification: the AI features your app needs, the agents and tools behind them, and how they fit together. Then through structured development phases and a production-level deployment plan ready for Claude Code, Hermes, Kiro, Antigravity, or any AI coding agent.
Not just vibe coding, but production-level development. It's way more powerful than /plan.

Get Started → See How It Works

"You've made something really really really cool. I'm almost done with our driver app. Will be field tested by Friday. I don't think I would have built what this is going to become without it."
— Wihan Booyse, Kriterion.ai

Get Started

Two commands. Runs on your machine.

install & launch

$ uv tool install spec4 --refresh

$ spec4

You'll need Python 3.12+, uv, and an API key for any supported provider — Anthropic, AWS Bedrock, Cohere, Google Gemini, Mistral, Nebius, or OpenAI. Add a search provider key — Tavily and Exa are currently supported — if you want the agents to ground their recommendations with web search (optional). Spec4 runs locally at localhost:8050 — your code and your keys never leave your machine. Prefer source? Clone the repo and run make spec4.

localhost:8050

Spec4 running locally — the app landing page

Proof, Not Promises

Spec4, applied to itself.

Built with Spec4 (BWS4) is a living showcase of common AI application patterns — from embeddings up to planning agents and multi-agent collaboration — presented as a collection of small, self-contained example apps.

Every one of them was planned and specced with Spec4 — the showcase's own .spec4/v0/ round ships in its repo, so you can read the spec behind what you're seeing.

Explore the showcase →

The Pipeline

See how it works.

🔍 CodeScanner → 🧠 Brainstormer → 🤖 Agentifier → 🎨 Designer → ⚙️ StackAdvisor → 📋 Phaser → 🚀 Deployer

Walk through each step of the Spec4 pipeline. Learn how CodeScanner reads your codebase, how Brainstormer defines your vision, how Agentifier specs the AI features your app needs, how Designer mocks your UI, how StackAdvisor recommends your tech stack, how Phaser breaks it into executable phases, and how Deployer plans your path to production.

See the full pipeline →

The Problem

Great ideas deserve great specs.

Agentic coding tools like Claude Code, Hermes, Kiro, and Antigravity have transformed how developers write software. But they can only be as good as the plan they're given. A vague or incomplete prompt leads to a vague or incomplete result — and an expensive amount of back-and-forth. And if your app should have AI inside it, most plans never even ask.

Spec4 bridges the gap between "I have an idea" and "I have a production-ready implementation plan." It structures your thinking, fills the gaps you didn't know existed, and produces machine-readable specs that agentic coders can execute reliably — phase by phase.

How It Works

A pipeline built for the way developers actually work.

🔍 CodeScanner → 🧠 Brainstormer → 🤖 Agentifier → 🎨 Designer → ⚙️ StackAdvisor → 📋 Phaser → 🚀 Deployer

01 🔍 CodeScanner (optional)

Reads your existing codebase so nothing gets lost in the transition to what's next.

code_review.json

02 🧠 Brainstormer

Collaborates with you to define a clear, complete vision — with a behavioral spec for every feature.

vision.json + feature_specs.json

03 🤖 Agentifier

Identifies every AI opportunity in your vision, right-sizes each one, and drafts a full implementation spec per feature.

ai_features.json

04 🎨 Designer (optional)

Generates a self-contained HTML mock of your UI so you have a visual target before a line of code is written.

design/mock.html

05 ⚙️ StackAdvisor

Guides you through selecting a technology stack that fits the vision and your requirements.

stack.json

06 📋 Phaser

Breaks the build into executable phases designed for first-attempt success.

phases.zip

07 🚀 Deployer

Plans your path to production — coding agent guidance, infrastructure, CI/CD, and monitoring.

deployment-plan.md

See the full pipeline

Beyond the First Draft

Built for real development.

Versioned rounds

Every planning round is self-contained in .spec4/v0/, v1/, v2/… When your coding agent finishes a round, Spec4 opens the next one — a repeatable, version-over-version development loop, not a one-shot plan.

Staleness detection

Every agent shows its state at a glance — Start, Modify, Needs Update, Required — driven by the dependency graph between artifacts. Change your vision, and Spec4 knows exactly which downstream specs need revisiting.

⏩ Fast Forward

In a hurry? Fast Forward sweeps the remaining decisions with the agent's best recommendations, then presents the complete set for your review. You stay in control — just faster.

Jump in anywhere

Start at any agent, revise any artifact, and pick up where you left off. Every agent supports revision rounds that carry your earlier decisions forward.

Local & private

Spec4 runs on your machine. Bring your own keys for Anthropic, AWS Bedrock, Cohere, Google Gemini, Mistral, Nebius, or OpenAI — they're stored only in your browser, and your code never leaves your computer.

Greenfield or brownfield

Starting fresh or evolving an existing codebase — Spec4 asks which, scans what's already there, and plans the next round on top of reality, not assumptions.

See It in Action

From one paragraph to a production plan.

Here's how Spec4 handled a real project brief:

project brief

> "I'd like to create an application to help healthcare professionals in regions where limited resources are available to be as efficient as possible — providing knowledge, advice, and helping to streamline processes and paperwork."

From this single paragraph, Spec4's pipeline clarified the target users, identified missing decisions (offline capability? local language support? integration with existing records systems?), recommended a technology stack, and produced a phased implementation plan — all ready to hand directly to an agentic coding tool.

Walk through the example

Stop prompting. Start speccing.

Spec4 is free and open source. Bring your own AI provider and model.

Get Started → How It Works