convez.ai
AgenteAutomaçãoAvançado
nt-nerdtechnic

Navide: Orquestrador de Agentes Multi-IA

nt-nerdtechnic/Navide

Coordene múltiplos agentes de IA em pipeline SDLC automatizado

Ver no GitHub
BomPluginScore · de 0 a 100

Navide transforma múltiplos agentes de IA (Claude, Codex, Gemini) em um time coordenado de engenharia, executando tarefas em paralelo através de um pipeline SDLC com 4 estágios (Requirements → Design → Implementation → Testing). Inclui orquestração automática de handoffs, análise local com LLM e editor nativo para workflows de IA.

Instalação

$npx convez-ai nt-nerdtechnic/Navide

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
44
Forks
5
Acompanhando
1
Saúde
100%
Último commit
há 5 horas
Versão
v0.2.44
Issues abertas
19
Contribuidores
1

Casos de uso

  • Automação de ciclos completos de desenvolvimento (requisitos até testes)
  • Execução paralela de agentes especializados (frontend, backend, QA)
  • Orquestração de fluxos complexos com handoffs automáticos entre estágios
  • Análise e interpretação de intent de agentes com LLM local

Veredito

Ferramenta ambiciosa para quem quer ir além de agentes isolados, coordenando múltiplos LLMs em pipelines estruturados. Requer setup local robusto (Node, Python, Ollama) e é ideal para equipes técnicas que buscam automação de processos SDLC completos, não apenas chatbot.

README

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

🌌 Navide (Agent-Team)

Stop chatting with an agent. Start leading a team. Run Claude Code, Codex, and Gemini CLI simultaneously — orchestrated across an SDLC pipeline with automated handoffs and multi-agent coordination.

English | 繁體中文

Electron Vue 3 Python Platform License


⚡ Why Navide (Agent-Team)?

A single AI coding agent has its limits — complex tasks require constant waiting, context is finite, and a single model often struggles to switch between PM, Backend, and QA mindsets.

Navide transforms your AI tools from isolated chat boxes into a coordinated engineering team.

Pain pointNavide's Solution
Linear WaitingParallel Execution: Run multiple agents (Frontend/Backend) simultaneously within a Stage.
Context LossAutomated Handoffs: Context is extracted and injected across 4 SDLC stages automatically.
Stuck AgentsLocal LLM Analyzer: A dedicated local LLM interprets agent intent to keep the pipeline moving.
Manual InputFull Auto Mode: AI answers the agent's questions based on your task context.

🚀 Core Features

🛠️ 4-Stage SDLC Pipeline

Transform a simple task description into a finished product through a fully automated pipeline: Requirements → Design → Implementation → Testing. Navide picks the right role and agent for every step.

🤝 Manager Coordination Mode

Designate a Manager Agent (e.g., Claude) to orchestrate a team of Worker Agents.

  • Manager dispatches instructions via ---DISPATCH---.
  • Workers report progress or ask questions via ---ASK---.
  • Manager decides when the stage is complete.

🧠 Local LLM Perception (Analyzer)

Beyond simple text matching, Navide uses Ollama or llama.cpp to interpret the CLI's intent in real-time.

  • Intent Detection: Recognizes if the agent is asking a question or is stuck.
  • Full Auto: Automatically answers technical questions to achieve 100% autonomy.

✍️ AI-Native Editor & Explorer

A built-in editor designed specifically for AI workflows:

  • AI Hunks: Review AI-proposed edits with inline diffs and per-hunk acceptance.
  • Ghost Text: Inline completions as you type.
  • Cmd+K Rewrite: Select code and give instructions for instant AI refactoring.

📊 Real-Time Token Tracking

Track costs accurately by parsing log files directly from the providers. No API keys required for tracking — categorized by Stage and Run.


🏁 Quick Start

1. Prerequisites

Navide (Agent-Team) is a local-first tool. Ensure you have the basics:

  • Node.js 22+ & pnpm 10+
  • Python 3.12+ & uv 0.11+
  • macOS 13+
  • (Optional) Ollama for Local Analysis.

2. Installation

# Clone the repository
git clone https://github.com/nt-nerdtechnic/Agent-Team
cd Agent-Team

# Install dependencies
pnpm install
uv --project backend sync

3. Run

pnpm dev

Upon launch, the Onboarding Wizard will guide you through the setup and tool detection.


🏗️ Architecture

Navide is built with a dual-engine design:

  • Orchestration Engine (Electron + Vue 3): Manages UI, state machines, and terminal grids.
  • Perception Engine (Python FastAPI): Handles PTY management, real-time log reading, and local LLM inference.
┌───────────────────────────┐      ┌───────────────────────────┐
│     Electron/Vue UI       │      │   Python FastAPI Backend  │
│ (Orchestrator & Terminal)  │ <──> │ (PTY, Logs, LLM Analyzer) │
└─────────────┬─────────────┘      └─────────────┬─────────────┘
              │                                  │
              ▼                                  ▼
      ┌──────────────────────────────────────────────────┐
      │  External Agent CLIs (Claude, Codex, Gemini)      │
      └──────────────────────────────────────────────────┘

🔒 Safety & Privacy

  • 100% Local: All computation and orchestration data stay on your machine.
  • No Telemetry: We don't track your tasks or your code.
  • YOLO Mode: Skip interactive confirmations at your own risk. YOLO mode grants agents unrestricted filesystem access.

🗺️ Roadmap

  • Git Preflight & Automated Task Branching
  • Cross-agent Route Engine (Enhanced Bus)
  • Windows & Linux Support
  • Support for more Agent CLIs (Aider, etc.)

📄 License

MIT © Navide Team

Ver o original no GitHub

Bom para

Relacionados