Relay — Quadros Interativos para Agentes IA
khanglvm/relay
Painéis visuais e formulários interativos para agentes IA em vez de ASCII
Ver no GitHubCLI que transforma a comunicação entre agentes IA e usuários, substituindo texto de terminal por painéis visuais no navegador. Permite que agentes façam perguntas estruturadas, apresentem gráficos, diagramas e protótipos interativos, recebendo respostas e comentários como JSON. Zero dependências, funciona offline.
Instalação
npx convez-ai khanglvm/relay1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Agentes IA (Claude Code, Cursor, Codex) fazendo perguntas estruturadas com formulários
- Visualização de arquiteturas, fluxos e dados via Mermaid, Graphviz ou PlantUML
- Prototipagem e feedback em tempo real com comentários inline em gráficos e tabelas
Veredito

Relay resolve um problema real: agentes IA pedindo input apenas via terminal. A implementação é sólida (zero deps, offline, autosave), integração com múltiplos agentes é clara, e o fluxo usuário é intuitivo. Recomendado para quem trabalha com Claude Code ou Cursor em contextos que exigem feedback visual estruturado.
README
A tradução para o português deste README ainda está na fila.
relay
The relay between AI agents and humans — browser boards instead of terminal walls.
rly is a CLI that lets your AI agent ask questions and present work in a
local browser board — real forms, charts, diagrams, sortable tables, images,
and live prototypes instead of terminal text and ASCII art. You click, comment
on any element, and Submit; the agent gets your answers and comments back as
JSON and keeps working.
See it
The whole loop — your agent asks in the terminal, a board opens in your browser; you pick a style, expand the chart, comment on a data point, leave a note, Submit — and the agent picks up your answers and keeps going:

Quick start
# the rly CLI
npm i -g @khanglvm/relay
# the agent skill → every detected agent (Claude Code, Codex, Cursor, …)
npx skills add khanglvm/relay --skill relay --all
# enforce it (recommended) — a skill alone is an ignorable hint; append relay's
# short usage rules to your agent's global instructions (per-agent commands below)
rly skill rules
That's it. Next time your agent needs a decision or wants to show you a plan, it opens a board like the ones above and waits for your Submit.
Tell your agent to actually use it
rly skill rules prints relay's short usage rules. A skill is just an ignorable
hint — these rules in the file your agent always reads are what make it reach for
relay. Append them to your agent's global instructions:
rly skill rules >> ~/.claude/CLAUDE.md # Claude Code
rly skill rules >> ~/.codex/AGENTS.md # Codex
rly skill rules >> ~/.gemini/GEMINI.md # Gemini CLI
rly skill rules >> ~/.config/opencode/AGENTS.md # OpenCode
Cursor and GitHub Copilot keep global rules in a settings panel, not a file — run
rly skill rules and paste the output into Cursor's Settings → Rules → User Rules
or Copilot's custom-instructions.
rly skill rules >> ~/.codeium/windsurf/memories/global_rules.md # Windsurf
rly skill rules >> ~/.factory/AGENTS.md # Droid (Factory)
Any other agent: run rly skill rules and paste the block into whatever file or
settings panel it reads as global instructions.
What it improves
| Without relay | With relay |
|---|---|
| Six questions asked one at a time in the terminal | One board, all questions, real form controls, optional "Other" + notes |
| "Option B is the one with caching (see my last message)" | Each answer option carries its own image / chart / diagram — pick by looking |
| ASCII architecture art | Mermaid, Graphviz, PlantUML — zoomable, full-screen, even user-editable |
| Numbers buried in prose | Charts and sortable tables; screenshots and HTML prototypes in a sandbox |
| "Type done when finished reviewing" | A Submit button; answers, notes, and inline comments returned as JSON |
| Feedback = another wall of text | Click any chart point, diagram node, table cell, or sentence to comment — the agent replies and the thread grows on the board |
Everything autosaves in real time (drafts survive timeouts), multiple boards run at once, and the package has zero runtime dependencies — plain Node ≥ 18; Chart.js / Mermaid / Graphviz are vendored and lazy-loaded offline.
Learn more
rly help | every command at a glance |
rly agent | the full agent guide — spec format, all block types, annotations, patterns (docs/AGENT.md) |
rly schema | board spec JSON Schema |
| skills/relay/SKILL.md | the bundled skill |
Development
npm test # zero-dep smoke tests (spawns real servers, fake-submits)
Migration from quest-board
relay was formerly @khanglvm/quest-board (CLI: qbd) — that package is
deprecated. Storage moved from ~/.quest-board to ~/.relay (override with
RLY_HOME); legacy "html" / "htmlFile" spec fields keep working.
License
MIT © Le Vu Minh Khang