convez.ai
AgenteAutomaçãoAvançado
jayvee

Aigon

jayvee/aigon

Orquestre múltiplos agentes de IA em um mesmo codebase com specs e Kanban local

Ver no GitHub
BomPluginScore · de 0 a 100

Orquestrador spec-driven para múltiplos agentes de IA (Claude, Gemini, Codex, Cursor) que trabalham simultaneamente no mesmo repositório. Gerencia fluxo de trabalho através de Kanban local, CLI e dashboard web, mantendo specs em Markdown dentro do git sem vendor lock-in.

Instalação

$npx convez-ai jayvee/aigon

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
27
Forks
1
Acompanhando
0
Saúde
100%
Último commit
há 1 mês
Versão
—
Issues abertas
7
Contribuidores
10

Casos de uso

  • Coordenar trabalho simultâneo de múltiplos agentes IA em features
  • Implementar workflows autônomos com revisão e feedback em loop
  • Manter histórico de trabalho IA como specs versionadas em git

Veredito

Aigon é uma ferramenta madura para equipes que trabalham com múltiplos agentes IA simultaneamente, resolvendo o problema real de orquestração sem vendor lock-in. Requer Node ≥18 e familiaridade com git/CLI, mas oferece ROI significativo em workflows IA complexos.

README

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

Beta: Aigon is early-access software. APIs and config formats may change between releases. Use it, file issues, and expect rough edges. Stable release coming soon.

Aigon is an open source project by Sen Labs. Follow aigon.build for updates.


What is Aigon?

Aigon is a spec-driven harness for orchestrating multiple AI coding agents on the same codebase. Each feature is a Markdown spec in git; each agent runs in its own git worktree under tmux; the lifecycle moves through inbox → backlog → in-progress → review → done on a local Kanban board. Aigon doesn't try to be a model — it orchestrates a growing list of agent CLIs as engines.

You work with Aigon from wherever you already are: a web dashboard on localhost, slash commands inside your agent (Claude Code, Antigravity CLI, Codex CLI, Cursor, OpenCode, Kimi, Amp, GitHub Copilot CLI), or direct CLI commands in your terminal. All three surfaces read and write the same Markdown specs in your repo — no SaaS account, no vendor lock-in, no proprietary state.

It works with the AI coding tools you already use:

  • 🤖 Claude Code (cc)
  • 💎 Antigravity CLI (ag)
  • 📟 Codex CLI (cx)
  • 🎯 Cursor (cu)
  • 🧭 OpenCode (op)
  • 🌙 Kimi Code CLI (km)
  • ⚡ Amp (am)
  • 🐙 GitHub Copilot CLI (cp)

Mix and match. Aigon doesn't care which model writes the code — it just manages the workflow around them.

Why Aigon?

  • Bring your own subscriptions — use the AI products you already pay for (Claude, Codex, Antigravity, Cursor, OpenCode, Kimi, Amp, GitHub Copilot); no separate Aigon billing, no token packs, no markup.
  • Stay in the flow when a subscription lapses — switch agents or accounts mid-session instead of being stuck on one vendor's meter.
  • Spec-driven development you control — features and research live as Markdown specs that can be built, shipped, reset, and rebuilt for clean handoffs and retries.
  • Autonomous and multi-agent workflows — configure who drafts, implements, and reviews, then run long loops through a pre-built push/merge lifecycle.
  • Context that compounds — specs and implementation summaries feed forward into later work, so context accrues in the repo rather than evaporating with a session.
  • Transparent, local, team-aligned — agents run in plain tmux sessions you can attach to; CLI + git on your machine, no required hosted platform, and fits GitHub PR review cleanly.

Quick start

Prerequisites: Node.js 18+, Git 2.20+, and tmux (for Fleet/worktree mode). See the Getting Started guide for platform-specific install instructions.

# Install aigon globally (beta)
npm install -g @senlabsai/aigon@next

# First-time setup — installs agent CLIs, configures terminal,
# starts the dashboard. Auto-launches on first aigon command, or
# run it explicitly:
aigon setup

# Bring aigon into your project
cd /path/to/your/project
aigon apply                            # Bootstraps .aigon/, kanban folders, hooks
aigon install-agent cc                 # Install the Claude Code agent (others: ag, cx, cu, op, km, am, cp)
aigon doctor                           # Verify environment

Aigon gives you three surfaces over the same Markdown specs in your repo. Pick the one that fits the moment — the state is identical across all three.

1. Dashboard — the recommended starting point

The dashboard is the daily command center. It's always-on after aigon setup and gives you the Kanban view of every feature, research topic, and piece of feedback across its lifecycle:

aigon server start                     # Starts at http://localhost:4100

Open the dashboard, click "New Feature", write a one-line description, and drag the card from inbox → backlog → in-progress. Click an agent button (cc, cx, cu, …) and a worktree spins up with a live tmux session you can attach to. Watch progress in the Monitor tab, review the diff in the in-evaluation column, and merge with one click. No commands to memorise.

The Reports tab shows cycle time, commits per feature, and rework ratio across your whole pipeline. Telemetry breaks down token spend per agent and activity. Open it on your phone over LAN or Tailscale to check in from the couch.

2. Slash commands — when you're already inside an agent

If you're already in Claude Code, Antigravity CLI, Codex, Cursor, OpenCode, Kimi, Amp, or GitHub Copilot CLI, drive Aigon from where you are. The fastest path is /aigon:feature-now:

/aigon:feature-now dark-mode
Add a dark mode toggle with system preference detection, persist
the choice in localStorage, and default to the system preference.

That one command creates the spec, assigns an ID, creates a feature branch, and starts implementing — all in your current repo, using your default agent. You stay in the agent the whole time.

3. CLI — terminal-native flow

For terminal-first workflows or scripting, drive Aigon by command. feature-create writes a bare spec from your description and launches Claude Code (cc) interactively with the drafting context pre-loaded:

aigon feature-create dark-mode --agent cc "dark mode toggle with system preference detection, persisted in localStorage"

You have a real conversation with the agent — it asks clarifying questions, proposes user stories and acceptance criteria, you push back and iterate, and when you're satisfied you exit the session. The drafted spec lands in docs/specs/features/01-inbox/ for a final review. When you're happy with the draft, move it forward:

aigon feature-prioritise dark-mode           # Assigns an ID, moves to backlog
aigon feature-start dark-mode cc             # Single-agent Drive mode in a worktree
# or
aigon feature-start dark-mode cc cx          # Fleet mode — Claude and Codex race the implementation
aigon feature-eval dark-mode                 # Compare results, pick the winner

Demo

The Aigon dashboard shows features across their full lifecycle — inbox → backlog → in-progress → in-evaluation → done — with live agent session status, commit activity, telemetry, and the repo-wide All Items index for features, research, and feedback.

Documentation

Full documentation lives at aigon.build/docs:

  • 📘 Getting Started — install and run your first feature
  • 🧭 Core Concepts — workflows, execution modes, reliability
  • 🛠 Guides — Drive, Fleet, Autopilot, Research, Dashboard, Telemetry
  • 📚 CLI Reference — every command documented
  • 🔍 Agents — Claude, Antigravity, Codex, Cursor, OpenCode, Kimi, Amp, GitHub Copilot setup

Community and support

  • 💬 GitHub Discussions — questions, workflows, design ideas
  • 🐛 Issues — bugs and concrete feature requests
  • 📖 Docs — the full manual
  • 🧑‍💻 Contributing — how to set up a dev environment and submit PRs
  • 🔒 Security policy — how to report vulnerabilities privately
  • 📰 aigon.build — follow here for launch updates and release notes
  • 🏢 Sen Labs — the company behind Aigon

License

Apache License 2.0 — Copyright 2025–2026 Sen Labs

Ver o original no GitHub

Bom para

Relacionados