How It Works
Agents CodeScanner Brainstormer Agentifier Designer StackAdvisor Phaser Deployer
BWS4 About Get Started →
About

About Spec4

What Spec4 is

Spec4 AI is a project planning tool for developers who are already using — or planning to use — agentic coding tools for spec-driven development. It's a free, open-source app that runs locally on your machine: you bring your own AI provider and keys, and your code never leaves your computer.

It's built on a simple observation: the quality of what an agentic coding tool produces is almost entirely determined by the quality of the spec it's given. Vague specs produce vague code. Incomplete specs produce incomplete code. Specs that grow organically as the conversation progresses lead to drift, contradiction, and rework. And as apps increasingly have AI inside them, the spec has to cover that too — which features use AI, how ambitious each one should be, and what it takes to run them in production.

Spec4 makes the spec the first-class artifact. Seven specialized agents — each with a distinct role and a clear handoff — work together to build a spec that includes the AI inside your app, a UI mock, and a deployment plan: precise, complete, and structured exactly the way a coding agent needs them.

The pipeline at a glance

Agent Role Output
CodeScanner (optional) Reads an existing codebase and confirms the current state code_review.json
Brainstormer Collaborates on the project vision through structured dialogue, with a behavioral spec per feature vision.json + feature_specs.json
Agentifier Identifies AI opportunities, assigns complexity tiers, and drafts per-feature AI implementation specs ai_features.json
Designer (optional) Generates a self-contained HTML mock of the application UI from a structured design manifest design/mock.html + design/manifest.json
StackAdvisor Guides technology stack selection with trade-off analysis stack.json
Phaser Breaks the build into phase-by-phase implementation prompts phases/phase<N>.md (phases.zip)
Deployer Plans the deployment strategy and coding agent workflow deployment-plan.md

Design principles

One question at a time.
Every agent in Spec4 asks one focused question and waits for your answer. No overwhelming forms, no multi-part questions, no assumptions. The pace is yours.
Confirmed, not assumed.
Every finding is confirmed with you before it moves forward. Every technical standard is verified against canonical documentation. Every dependency is added to the spec only after you approve it.
Bring your own everything.
Spec4 is provider-agnostic. Choose the AI model that fits your budget and preference. Use whatever agentic coding tool you work with. The output is plain JSON and Markdown — yours to use however you like.
Downstream quality starts upstream.
Each agent's output is designed to feed the next. A thorough code review leads to better stack recommendations. A complete vision leads to more accurate phases. The pipeline is only as good as each handoff.
Right-size the AI.
Not every feature needs an agent — and some don't need AI at all. Agentifier recommends the cheapest tier of complexity that genuinely does the job, and says so plainly when deterministic code is the right answer.
Nothing leaves your machine.
Spec4 runs locally. Your API keys live only in your browser, your artifacts save to your project directory, and there's no telemetry, no hosted service, and no account.

Who Spec4 is for

Spec4 is for developers who are already using — or planning to use — agentic coding tools, and who want to get more reliable, consistent results from them. It's particularly useful when:

The project is complex enough that a single prompt won't cut it
You're working on an existing codebase and need to plan a significant change
You want a repeatable, structured process rather than ad-hoc prompting
You're working with a team and need a shared, documented plan before implementation begins
You're adding AI or LLM features and want them specified — tiers, evals, budgets — not improvised mid-build

Open source

Spec4 is free to use, open source under the Apache 2.0 license. Source code, contribution guidelines, and the roadmap are available on GitHub — the Agentifier pattern library is a particularly good place to contribute.

Want proof it works? Built with Spec4 (BWS4) is a live showcase of common AI application patterns — planned and specced with Spec4 itself, with the full planning round in its repo.

View on GitHub

Stop prompting. Start speccing.

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

Get Started → How It Works