Built with Spec4

See what a Spec4 spec becomes.

Built with Spec4 (BWS4) is a live showcase of common AI application patterns, presented as small, self-contained example apps. Each one demonstrates a pattern end-to-end — how it's built, what it depends on, and how it behaves — so you understand it by seeing it work, not by reading about it abstractly. And every one of them was planned and specced with Spec4 itself.

Visit the showcase → View the repo

The example apps

The gallery walks up Agentifier's complexity ladder — embeddings, single call, RAG, tool use, chained calls, and more, each live and self-contained. Here are two from the top of the ladder:

Planning agent

A planner call decomposes a trip-day goal into a visible plan of research and synthesis steps — and because the plan is produced and displayed before execution, you can inspect exactly what the agent intends to do, and nothing runs until you say so. On your go-ahead, the agent executes the research steps with a live web-search tool and composes a one-day itinerary.
Run it live →

Multi-agent collaboration

A buyer agent negotiates with two rival seller agents that hold private constraints neither can see. The agents are peers, not workers — each with its own goals and private information, communicating only through explicit sender→recipient messages, with opacity enforced by the plumbing rather than by asking a model nicely. Modeled on the A2A protocol's data model — and when the deal closes, you unseal every hidden position.
Run it live →

The rest of the gallery — embeddings, single call, RAG, tool use, chained calls, and orchestrated subagents — is live too, and the gallery is designed to grow. Have you written something amazing with Spec4? Let us know and we may add it to BWS4!

Specced by Spec4, naturally

BWS4 exists to answer a fair question: what does a Spec4 spec actually produce? The showcase's own planning round — vision, feature specs, AI feature catalog, stack spec, development phases, and deployment plan — ships in the repo under .spec4/v0/. You can read the spec, then look at the running code it produced.

The examples map directly onto Agentifier's nine-tier ladder: the planning agent is tier 7, the peer-to-peer collaboration is tier 9, and the rest of the gallery covers the rungs below — each one right-sized, exactly as Agentifier would spec it.

Under the hood

BWS4 is a real production deployment, not a slide deck — and it runs entirely within free usage tiers, so it's easy to fork, run locally, and deploy without a billing surprise.

Frontend

TypeScript, React Router, Tailwind CSS, and Vite — with per-example code-splitting, so each app is lazy-loaded independently and the gallery can grow without slowing the landing page.

Backend

Python 3.12 with FastAPI and SQLAlchemy. Shared framework services — generation, embeddings, storage — are reused across example apps, so each new pattern only adds what's unique to it.

Models & data

LiteLLM routing across OpenRouter and Groq free-tier model chains, in-process sentence-transformers embeddings, and Neon serverless Postgres with pgvector as both the relational store and the vector index.

Operations

Deployed on Render free tiers from a checked-in blueprint, with Sentry error tracking and model-call observability that no-ops cleanly when unconfigured — the deployment story straight out of its Spec4 deployment plan.

See what a Spec4 spec becomes.

The example apps

Specced by Spec4, naturally

Under the hood

See it work, then read the spec.