convez.ai
AgenteAutomaçãoAvançado
Ayont

ayontclaudian — Plugin Obsidian Multi-Agent

Ayont/ayontclaudian

12 agentes de IA em um sidebar Obsidian com histórico sincronizado

Ver no GitHub
BomPluginScore · de 0 a 100

Plugin Obsidian que integra 12 agentes de código (Claude, Codex, Antigravity, Kimi e outros) com seletor unificado de modelo e troca de provedor durante a conversa. Transforma seu vault em workspace de desenvolvimento, permitindo que os agentes leiam, editem notas e executem comandos shell nativamente.

Instalação

$npx convez-ai Ayont/ayontclaudian

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
4
Forks
1
Acompanhando
0
Saúde
57%
Último commit
há 11 horas
Versão
5.144.4
Issues abertas
11
Contribuidores
23

Casos de uso

  • Desenvolvimento assistido por IA sem sair do Obsidian
  • Comparação e teste de múltiplos agentes na mesma sessão
  • Rastreamento de tokens, custos e rate-limits por provedor
  • Automação de tarefas de código com MCP e skills customizados
  • Verificação e retry automático de objetivos com loops de goal

Veredito

Ferramenta robusta e madura para quem vive no Obsidian e quer múltiplos agentes sempre à mão. A arquitetura provider-nativa (não simulada) e o suporte a troca mid-chat sem perder histórico destacam. Requer familiaridade com CLIs de agentes e setup de APIs.

README

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

ayontclaudian

GitHub stars GitHub release Tests License

Twelve AI coding agents. One Obsidian sidebar. Your vault is the workspace.

ayontclaudian hero banner

Claude Code · OpenAI Codex · Antigravity · Kimi · Mistral Vibe · Grok · Opencode · Hermes · Cline · Pi · DSH


Why

Every coding agent ships its own terminal UI, its own session store, its own idea of project context. ayontclaudian embeds the real CLIs as chat runtimes inside Obsidian — so the agent reads and edits your notes, searches your vault, runs shell commands and git workflows where your work actually lives. Switch providers or models mid-conversation without losing a single message.

  • 🔌 Provider-native, not simulated — each adaptor drives the official CLI or SDK (Claude Agent SDK, codex app-server JSON-RPC, ACP for Opencode/Hermes). Real streaming, real tool calls, real session resumes.
  • 🧭 Mid-chat provider switching — fork a conversation to Codex when Claude gets stuck; history stays visible, state stays per-provider.
  • 📊 Usage and cost center — token spend per provider/model with subscription-aware pricing that never invents a rate.
  • ⏳ Rate-limit chips in the status bar — live windows like 5h: 34% · Reset 1h 30m, read straight from the provider's own reporting (Codex) plus local rolling-window tracking for everyone else.
  • 🤖 Multi-agent missions — a master prompter writes one tailored prompt per specialist, then routes subtasks to whichever provider still has capacity headroom.
  • 🎯 Goal loops — /goal runs verify-and-retry rounds with an adversarial verifier until the objective actually holds.
  • ↩️ Undo everything — vault snapshots back every turn; rewind and fork supported natively (Claude) or via transcript replay (Codex).
  • ✏️ Inline edit — select text anywhere in a note, hotkey, word-level diff preview, accept or reject.
  • 🧩 MCP, skills, slash commands, mentions — vault-scoped skills, managed MCP servers, reusable prompt templates, file and subagent mentions.
  • 🛡️ Plan mode — let the agent explore first and present a plan for approval before it touches anything.

Providers

ProviderTransportHighlights
Claude CodeOfficial Agent SDKFull feature set: rewind/fork, native history, vault MCP, plan mode
OpenAI Codexapp-server JSON-RPCFork, JSONL history reload, subagents, images, native rate-limit windows
Antigravityagy --printSingle-shot with transcript-tailed state recovery
KimiNDJSON + ACPMoonshot coding agent
VibeNDJSONMistral coding agent
GrokDelta JSONxAI, resumable sessions
OpencodeACPShared Agent-Client-Protocol transport
HermesACPModel catalog plus modes from session/new, SQLite state history
Cline--printHome of the /goal verification loop
Pi--printMinimal, fast
DSH--printReads model selection from DeepSeek Harness config
Grok (App) · PerplexityDesktop relayExperimental: drives the installed desktop app's own chat window. No API key, no automatic vault recall, off by default

Adding the next provider is mechanical — see the provider docs in this repo.

Install

BRAT (recommended)

  1. Install BRAT
  2. Choose Add beta plugin and enter Ayont/ayontclaudian
  3. Enable ayontclaudian in community plugins

Manual

Grab main.js, manifest.json and styles.css from the latest release into .obsidian/plugins/realclaudian/.

Requirements

Obsidian desktop plus at least one provider CLI installed and logged in. Missing CLIs can be installed directly from the plugin settings — including standalone updaters for opencode, grok and claude.

Quick start

  1. Open the Claudian sidebar (ribbon icon)
  2. Pick a provider and model in the unified picker
  3. Ask something about your vault — try "summarize everything I wrote about project X last week"
  4. Watch the status bar: context usage on one side, rate-limit windows on the other

Status bar at a glance

● Claude   bereit   42%  [####......]   5h: 34% - Reset 1h 30m   7T: 8% - Reset 4d 2h
  provider  state   context meter    rate-limit windows (native Codex data / local tracker)

Development

npm install
npm run dev        # watch build
npm test           # 7400+ unit and integration tests
npm run build      # production bundle

Architecture notes live in CLAUDE.md and per-directory guides (src/core/CLAUDE.md, src/features/chat/CLAUDE.md, and more). Every behavior change lands test-first; provider claims must be verified against the real binary, never guessed.

How this fork differs

Forked from YishenTu/claudian and since rebuilt around a provider-neutral core: twelve adaptors behind one runtime contract, a budget and capacity engine for multi-agent routing, a cost center with honest pricing, CLI auto-update management, and a test suite that pins every trap we have ever hit. Not on the Obsidian community registry — updates ship via BRAT and GitHub releases.

Acknowledgments

  • YishenTu/claudian — the original plugin
  • Anthropic, OpenAI, Moonshot, Mistral, xAI — the agents this embeds
  • The Obsidian team for a canvas worth automating

If ayontclaudian saves you a tab-switch today, a star helps other vault people find it.

Ver o original no GitHub

Bom para

Relacionados