Skip to content

About

Portuguese-language copilot that turns a business idea into real-world tests: a guided conversation, a living canvas and evidence from what actually happens.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

create — wordmark with the spark symbol and three connected plan cards

Getting started · Usage · Português

License: MIT Node.js 24 React 19 TypeScript

The create workspace: the conversation on the left and the plan canvas on the right, with the Now card of the fictional Chabor demo

From a business idea to its first real evidence, one small test at a time.

create is a copilot for people starting their first business, in Brazilian Portuguese. You tell it your idea; a short interview turns it into a visual plan of cards and points to one thing to do now. Then you run that test in the real world, come back with what happened, and the plan proposes a revision you can accept or reject.

Start small, learn for real. Hypotheses, research and evidence keep their origin, and a ticked checklist is never treated as proof that people want what you are building.

Features

  • From conversation to canvas. The chat asks one question at a time and adds cards to an editable canvas while it works; each card is saved before it appears.
  • One next step. At most one action sits in Agora (Now), and prerequisite links between cards reject cycles and duplicates.
  • Your edits stay. Titles, blocks, checklists, positions and links you change are preserved; AI changes that would overwrite them become proposals.
  • Research with sources. Optional web search adds links, excerpts and consultation dates. When search is off, the app says so instead of inventing results.
  • Learn from real tests. Log what happened and compare the proposed revision side by side before applying it.
  • No sign-up. Projects belong to an anonymous browser session (up to three), with JSON export and import.

Getting started

Note

You need Node.js 24 and npm. No API keys are required to try it locally: without them, projects are kept in memory and the assistant runs in a basic guided mode, clearly labeled as such. The interface is in Brazilian Portuguese.

git clone https://github.com/PHMFG/create-dia-zero.git
cd create-dia-zero
npm ci
cp .env.example .env.local
npm run dev

Open http://127.0.0.1:5173. The same process serves the Vite frontend and the Hono API; restart it after changing backend code.

npm test
npm run typecheck
npm run build
Optional configuration

Add keys to .env.local to turn on each service. They are read by the server only.

Variables What they enable
SUPABASE_URL, SUPABASE_PUBLISHABLE_KEY Persistent projects with anonymous auth and row-level security. Required in production.
OPENCODE_GO_API_KEY, OPENCODE_MODEL, OPENCODE_BASE_URL Generated replies through an OpenAI-compatible endpoint (OpenCode Go by default).
TYPESAFE_API_KEY, TYPESAFE_MODEL Message routing with Jev; simple rules are used without it.
TAVILY_API_KEY Web research.

To use your own Supabase project, apply supabase/schema.sql and then supabase/cancellation.sql, enable Anonymous Sign-Ins under Authentication → Sign In / Providers, and keep the private schema out of the Data API. vercel.json deploys the static build together with the API as a Node.js function.

Usage

  1. Run npm run dev and open the app.
  2. Click Explore um projeto de exemplo to open Chabor, a fictional matcha brand used as a guided demo, or type an idea such as “Quero abrir uma cafeteria diferente na minha cidade.”
  3. On the canvas, the Agora card is your next step. Tick its checklist, open Sobre for the full card, or use the chat bubble to ask a question about that card.
  4. After running the test for real, click Experiência, describe what you observed and review the proposed revision.

Ticking two steps of the Now card on the canvas, opening the card, ticking a third step and closing it

Screens

Show the gallery

Landing page with the idea field, three suggestions and a link to the example project
Start from an idea or open the example project.

Card details panel with view, edit and history tabs and a four-step checklist
Each card has a view, an editor and a history.

Revision dialog comparing the current plan with a proposed new card, with Keep my plan and Apply revision buttons
Logging an experience leads to a proposal you can apply or reject. Captured in the offline guided mode, which says so.

Two phone screens: the landing page and the plan in list view
On small screens, the conversation and the plan become tabs.

How it works

flowchart LR
  UI["Browser<br/>React + React Flow"] -->|message| API["Hono API"]
  API -->|"SSE: saved updates"| UI
  API --> DOMAIN["Shared domain rules"]
  API --> TOOLS["AI tools"]
  TOOLS --> MODEL["Generative model"]
  TOOLS --> ROUTER["Jev router"]
  TOOLS --> SEARCH["Tavily search"]
  API --> DB[("Supabase<br/>auth + RLS")]
  API -.->|local dev| MEMORY[("In-memory store")]
Loading

During a reply, the model can read the project, search, calculate and publish cards. The API checks every change against the shared domain rules and saves it with a version check before streaming it to the browser, which merges it with edits you have not saved yet. Stopping a reply is an explicit, durable request, so it also works behind a proxy. The architecture page of the wiki has the details (in Portuguese).

Roadmap

  • Conversation that builds an editable canvas with seven card types
  • Cards published one by one over SSE, each saved before it is shown
  • Human edits preserved, with conflicts turned into proposals
  • Web research with sources and consultation dates
  • Experience logging and revision proposals
  • Anonymous sessions with row-level security, plus JSON export and import
  • Offline guided mode and the Chabor demo project
  • Public demo back online (the deployment is paused)
  • Move projects between devices without exporting a file
  • Verify bot-protection tokens on the API
  • English interface
  • Continuous integration running the tests on Node.js 24

Documentation

The project wiki covers product, architecture, domain, AI, research, persistence, security and operations, and AGENTS.md explains how to continue development. Both are in Portuguese. Earlier verification reports are in docs/.

License and credits

The code is released under the MIT License. The create wordmark, the spark symbol and the images in docs/assets/ were made for this project; the screenshots show only fictional demo data.

Built with open-source software, each under its own license:

The optional services (OpenCode Go, Jev, Tavily, Supabase and Vercel) are not bundled and follow their own terms.

About

Portuguese-language copilot that turns a business idea into real-world tests: a guided conversation, a living canvas and evidence from what actually happens.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages