convez.ai
AgenteDesenvolvimento WebIntermediário
Vibisual

Vibisual

Vibisual/vibisual

Veja seus agentes IA pensarem — visualize e projete fluxos em um mapa de bolhas

Ver no GitHub
InicialPluginScore · de 0 a 100

Visualizador de agentes Claude Code em tempo real através de um mapa de bolhas interativo. Permite tanto monitorar o comportamento dos agentes via hooks quanto desenhar a orquestração de agentes visualmente em um canvas, substituindo configurações textuais por design gráfico.

Instalação

$npx convez-ai Vibisual/vibisual

1 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

Estrelas
1
Forks
1
Acompanhando
0
Saúde
57%
Último commit
há 1 dia
Versão
v0.2.8
Issues abertas
0
Contribuidores
2

Casos de uso

  • Debugar fluxos multi-agente em tempo real
  • Projetar orquestração de agentes sem tocar em JSON
  • Monitorar chamadas de ferramentas e handoffs entre agentes
  • Documentar visualmente a arquitetura de agentes

Veredito

Ferramenta promissora para quem trabalha com múltiplos agentes Claude Code. A visualização em tempo real e o design canvas reduzem significativamente a fricção operacional, embora o projeto esteja ainda em estágio bem inicial com testes limitados fora do Windows.

README

A tradução para o português deste README ainda está na fila.

Vibisual

See your AI agents think.

Visualize agent behavior through hooks, or build your harness by placing custom agent bubbles on a canvas.

License: Apache 2.0 Node.js Built with Claude Code Status: early


What it does

Vibisual does two things.

1. Visualizes Claude Code through hooks

Supported Claude Code hook events — including PreToolUse, PostToolUse, UserPromptSubmit, SessionStart, and others — become nodes on a live bubble map. Sub-agent spawns become edges. Tool calls become child bubbles. Keyword links connect related work across sessions.

The terminal output of a multi-agent Claude session is a tree printed as a wall of text. Vibisual draws that tree as it grows.

2. Designs the harness as a visual graph

The bubble map is both the runtime view and the design surface for your harness. Instead of editing settings.json in a text editor, you build the harness on a canvas:

  • Place agents as nodes. Drop a bubble onto the canvas to define a new sub-agent. Each node carries its own configuration — model, permission mode, tools, isolation, max turns, effort level, skills, and per-agent rules.
  • Wire them with edges. Connect agents with task edges to define handoffs and dependencies between them. The edges become the control-flow graph of your harness.
  • The graph defines the harness. In the current early release, Vibisual reads the graph and launches the corresponding Claude Code sub-agent workflow. The same canvas you designed on is the canvas you watch the run on.

What used to be a buried settings.json tree is now a workflow you can see, edit, and rearrange at any time.

Watch the full walkthrough

Vibisual — full walkthrough on YouTube

▶ Watch on YouTube — 2-minute walkthrough

Quick Start

Install on Windows

Vibisual is a hooks-based visualizer that runs on top of the Claude CLI. The Claude CLI must already be installed and available on your PATH.

  1. Download the latest installer from the Releases page and run it:

    Vibisual-0.1.0-setup.exe
    
  2. Launch Vibisual.

Build from source (contributors)

git clone https://github.com/Vibisual/vibisual.git
cd vibisual
pnpm install

# Build and launch the desktop app
node scripts/runapp.mjs

# Build a Windows installer
pnpm build:win

About the hook installer

On first launch, Vibisual writes a managed hook block into ~/.claude/settings.json so Claude CLI sessions can stream into the bubble map. A timestamped backup is kept next to it (.bak-vibisual-*). If you'd rather wire hooks yourself, set VIBISUAL_SKIP_HOOK_INSTALL=1 before first launch.

Tested on Windows. macOS and Linux builds are available but not extensively tested.

Status

Vibisual is currently an early preview release.

Version 0.1.0 is intended for experimentation, demos, and feedback. Expect bugs, incomplete features, rough edges, and occasional breaking changes.

Vibisual is not recommended for critical production workflows or repositories containing highly sensitive data.

Security and privacy

Vibisual uses Claude Code hooks to visualize agent activity. Hook payloads may include prompts, tool calls, file paths, shell commands, session metadata, and other local development context.

Vibisual is currently intended for local experimentation. Review the generated hook configuration before use, especially in repositories containing secrets, credentials, private code, customer data, or production systems.

Claude Code command hooks run with the permissions of your local user account. Only install and run hooks from code you trust.

License

Apache License 2.0 — see LICENSE. "Vibisual" and the Vibisual logo are trademarks of the project maintainers; see TRADEMARK.md for the policy.

Contributing

By contributing, you agree to the DCO sign-off requirement and the additional contribution terms described in CONTRIBUTING.md, including permission for project maintainers to relicense contributions for future commercial offerings.

Disclaimer

Vibisual is an independent open-source project and is not affiliated with, endorsed by, or sponsored by Anthropic.

Claude, Claude Code, and Anthropic are trademarks or registered trademarks of Anthropic, PBC. All product names, logos, and brands are property of their respective owners.

Ver o original no GitHub

Bom para

Relacionados