自律型AIコーディングアシスタント - サブエージェントオーケストレーション・Deep Research・MCP統合
Release Date: November 2, 2025
Codename: "Spectrum"
🎯 Blueprint Mode (Phase 1) - Complete hierarchical planning system
- Execution strategy (Single/Orchestrated/Competition)
- Budget management with cost estimation
- State persistence with checkpoint/resume
- Policy enforcement for security
🌍 Multi-Language /review - Code review in your native language
- 8 languages supported: Japanese, English, Chinese, Korean, French, German, Spanish, Portuguese
- AGENTS.md integration for automatic language detection
- Localized system prompts for each language
📊 Monitoring & Telemetry - Privacy-respecting analytics
- SHA-256 hashing for data protection
- Event tracking with structured logs
- Webhook integration (GitHub/Slack/HTTP, HMAC-SHA256)
🤖 Multi-LLM Support - Unified interface for all major providers
- OpenAI GPT-5-codex (Primary)
- Google Gemini 2.5 Pro/Flash (Search grounding)
- Anthropic Claude 3.5+ (Alternative)
- Local/Ollama models (Privacy-first)
| Feature | Description |
|---|---|
| 🤖 Auto-Orchestration | Automatically analyzes task complexity and coordinates specialized sub-agents in parallel |
| ⚡ 2.6x Faster | Parallel execution of independent tasks for maximum productivity |
| 🔍 Deep Research | Multi-source research with citations, contradiction detection (Gemini CLI + DuckDuckGo) |
| 🔒 Secure by Default | Sandbox isolation (Seatbelt/Landlock) with approval policies |
| 🌐 Multi-IDE Support | VS Code, Cursor, Windsurf extensions with unified experience |
| 🔌 MCP Integration | Bi-directional Model Context Protocol support (client & server) |
| 🌍 Cross-platform | Windows, macOS, Linux support with native performance |
📊 Interactive Mermaid Diagram (Click to expand)
graph TB
subgraph Client["🖥️ Client Layer"]
CLI["CLI<br/>codex-cli<br/>Cross-platform binaries"]
TUI["TUI<br/>ratatui-based<br/>Rich terminal UI"]
VSCode["VSCode Extension<br/>v1.0.0<br/>Blueprint UI<br/>Multi-language support"]
Cursor["Cursor IDE<br/>MCP Integration<br/>Composer support"]
WebGUI["Web GUI<br/>React + Vite<br/>Dashboard"]
end
subgraph Orchestration["🎯 Orchestration Layer (v1.0.0)"]
OrchestratorRPC["Orchestrator RPC Server<br/>16 RPC methods<br/>HMAC-SHA256 auth<br/>TCP/UDS/Named Pipe"]
ProtocolClient["Protocol Client<br/>@zapabob/codex-protocol-client<br/>TypeScript SDK<br/>React hooks"]
TaskQueue["Single-Writer Queue<br/>Concurrent task management<br/>Priority-based execution"]
LockManager["Lock Manager<br/>.codex/lock.json<br/>Repository-level concurrency"]
end
subgraph Core["⚙️ Core Runtime (Rust)"]
CoreEngine["Core Engine<br/>codex-core<br/>Session management<br/>Tool execution"]
Blueprint["Blueprint Mode<br/>v1.0.0 Complete<br/>3 execution strategies<br/>Budget & Policy enforcement"]
TokenBudget["Token Budget<br/>Thread-safe tracking<br/>Per-agent limits"]
AuditLog["Audit Logger<br/>Structured logs<br/>Security events"]
ProjectDoc["Project Doc<br/>AGENTS.md parser<br/>Language detection<br/>Multi-language support"]
end
subgraph Agents["🤖 Sub-Agent System"]
Supervisor["Supervisor<br/>Agent lifecycle<br/>Timeout: 5min<br/>Retry: 3x"]
CodeReviewer["Code Reviewer<br/>Multi-Language v1.0.0<br/>8 languages (JA/EN/ZH/KO/FR/DE/ES/PT)<br/>AGENTS.md integration<br/>ReviewLocale"]
TestGen["Test Generator<br/>Coverage 80%+<br/>Unit/Integration"]
SecAudit["Security Auditor<br/>OWASP Top 10<br/>CVE scanning"]
DeepResearch["Deep Researcher<br/>Multi-source<br/>Citation-based"]
CustomAgent["Custom Agents<br/>YAML-defined<br/>.codex/agents/"]
end
subgraph Research["🔍 Deep Research Engine"]
SearchProvider["Search Provider<br/>Cache TTL: 1h<br/>45x faster"]
GeminiCLI["Gemini CLI<br/>OAuth 2.0 PKCE<br/>Google Search Grounding"]
DuckDuckGo["DuckDuckGo<br/>Zero-cost<br/>No API key"]
Citation["Citation Manager<br/>Source tracking<br/>Confidence scoring"]
end
subgraph MCP["🔌 MCP Integration (15+ Servers)"]
CodexMCP["codex mcp-server<br/>Self-hosted<br/>Rust-based"]
GeminiMCP["gemini-cli-mcp<br/>Google Search<br/>Gemini 2.5"]
ChromeMCP["chrome-devtools<br/>Browser automation"]
PlaywrightMCP["playwright<br/>E2E testing"]
SequentialMCP["sequential-thinking<br/>CoT reasoning"]
end
subgraph Storage["💾 Storage & Config"]
ConfigTOML["config.toml<br/>Model providers<br/>Sandbox settings"]
SessionDB["Session DB<br/>Conversation history<br/>Checkpoints"]
AgentDefs[".codex/agents/<br/>YAML definitions<br/>Agent configs"]
ArtifactArchive["Artifact Archive<br/>_docs/<br/>Implementation logs"]
BlueprintStore["Blueprint Store<br/>State persistence<br/>Checkpoint/resume"]
end
subgraph Monitoring["📊 Monitoring & Telemetry (v1.0.0)"]
Telemetry["Telemetry Module<br/>Privacy-respecting<br/>SHA-256 hashing<br/>Event tracking"]
Webhooks["Webhooks Module<br/>GitHub/Slack/HTTP<br/>HMAC-SHA256<br/>Real-time notifications"]
end
subgraph External["🌐 External Integrations"]
GitHub["GitHub API<br/>PR automation<br/>Issue management"]
Slack["Slack Webhooks<br/>Real-time notifications"]
CustomWebhook["Custom Webhooks<br/>HTTP POST endpoints"]
AudioNotif["Audio Notifications<br/>marisa_owattaze.wav<br/>Task completion sound"]
end
subgraph LLM["🤖 LLM Providers"]
OpenAI["OpenAI<br/>GPT-5-codex<br/>Primary provider"]
Gemini["Google Gemini<br/>2.5 Pro/Flash<br/>Search grounding"]
Anthropic["Anthropic<br/>Claude 3.5+<br/>Alternative"]
Local["Local/Ollama<br/>Llama models<br/>Privacy-first"]
end
CLI --> OrchestratorRPC
TUI --> OrchestratorRPC
VSCode --> OrchestratorRPC
Cursor --> CodexMCP
WebGUI --> ProtocolClient
ProtocolClient --> OrchestratorRPC
OrchestratorRPC --> TaskQueue
OrchestratorRPC --> LockManager
TaskQueue --> CoreEngine
CoreEngine --> Blueprint
CoreEngine --> TokenBudget
CoreEngine --> AuditLog
CoreEngine --> Supervisor
CoreEngine --> ProjectDoc
ProjectDoc --> CodeReviewer
Supervisor --> CodeReviewer
Supervisor --> TestGen
Supervisor --> SecAudit
Supervisor --> DeepResearch
Supervisor --> CustomAgent
DeepResearch --> SearchProvider
SearchProvider --> GeminiCLI
SearchProvider --> DuckDuckGo
SearchProvider --> Citation
CoreEngine --> CodexMCP
CodexMCP --> GeminiMCP
CodexMCP --> ChromeMCP
CodexMCP --> PlaywrightMCP
CodexMCP --> SequentialMCP
CoreEngine --> ConfigTOML
CoreEngine --> SessionDB
Supervisor --> AgentDefs
CoreEngine --> ArtifactArchive
Blueprint --> BlueprintStore
CoreEngine --> Telemetry
CoreEngine --> Webhooks
Telemetry --> GitHub
Webhooks --> Slack
Webhooks --> CustomWebhook
Supervisor --> AudioNotif
CoreEngine --> GitHub
CoreEngine --> Slack
CoreEngine --> CustomWebhook
CoreEngine --> OpenAI
GeminiCLI --> Gemini
CoreEngine --> Anthropic
CoreEngine --> Local
classDef clientClass fill:#e1f5ff,stroke:#01579b,stroke-width:2px
classDef orchClass fill:#fff9c4,stroke:#f57f17,stroke-width:3px
classDef coreClass fill:#ffebee,stroke:#c62828,stroke-width:2px
classDef agentClass fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
classDef researchClass fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px
classDef mcpClass fill:#fff3e0,stroke:#e65100,stroke-width:2px
classDef storageClass fill:#e0f2f1,stroke:#004d40,stroke-width:2px
classDef monitoringClass fill:#f1f8e9,stroke:#558b2f,stroke-width:2px
classDef externalClass fill:#fce4ec,stroke:#880e4f,stroke-width:2px
classDef llmClass fill:#ede7f6,stroke:#311b92,stroke-width:2px
class CLI,TUI,VSCode,Cursor,WebGUI clientClass
class OrchestratorRPC,ProtocolClient,TaskQueue,LockManager orchClass
class CoreEngine,Blueprint,TokenBudget,AuditLog,ProjectDoc coreClass
class Supervisor,CodeReviewer,TestGen,SecAudit,DeepResearch,CustomAgent agentClass
class SearchProvider,GeminiCLI,DuckDuckGo,Citation researchClass
class CodexMCP,GeminiMCP,ChromeMCP,PlaywrightMCP,SequentialMCP mcpClass
class ConfigTOML,SessionDB,AgentDefs,ArtifactArchive,BlueprintStore storageClass
class Telemetry,Webhooks monitoringClass
class GitHub,Slack,CustomWebhook,AudioNotif externalClass
class OpenAI,Gemini,Anthropic,Local llmClass
Interactive Mermaid diagram for GitHub viewers
📥 Download High-Resolution Diagram:
- SVG (Scalable Vector Graphics) - Best for web/print
- PNG (2400x1800px) - Best for presentations/social media
- Mermaid Source - Editable source code
Comprehensive system architecture diagram showing the complete v1.0.0 ecosystem with VSCode extension integration, orchestrator RPC layer, multi-language support, Blueprint Mode, and Monitoring & Telemetry (Updated 2025-11-02)
The Codex v1.0.0 architecture consists of 10 major layers with 55+ core components:
- 🖥️ Client Layer – CLI (codex-cli), TUI (ratatui), VSCode Extension (v1.0.0), Cursor IDE (MCP), Web GUI (React + Vite)
- 🎯 Orchestration Layer – Orchestrator RPC Server (16 methods, HMAC-SHA256), Protocol Client (TypeScript SDK), Task Queue (single-writer), Lock Manager (.codex/lock.json)
- ⚙️ Core Runtime – Core Engine (codex-core), Blueprint Mode (v1.0.0 Complete), Token Budget (thread-safe), Audit Logger (structured logs), Project Doc (AGENTS.md parser, Multi-language)
- 🤖 Sub-Agent System – Supervisor (agent lifecycle, 5min timeout, 3x retry), CodeReviewer (Multi-Language v1.0.0, 8 languages), TestGen, SecAudit, DeepResearch, Custom Agents (YAML-defined)
- 🔍 Deep Research Engine – Search Provider (cache TTL: 1h, 45x faster), Gemini CLI (OAuth 2.0 PKCE), DuckDuckGo (zero-cost), Citation Manager (confidence scoring)
- 🔌 MCP Integration – 15+ servers: codex mcp-server (Rust), gemini-cli-mcp (Google Search), chrome-devtools, playwright, sequential-thinking
- 💾 Storage & Config – config.toml (model providers, sandbox), Session DB (conversation history), Agent Definitions (.codex/agents/), Artifact Archive (_docs/), Blueprint Store (state persistence)
- 📊 Monitoring & Telemetry – Telemetry Module (privacy-respecting, SHA-256), Webhooks Module (GitHub/Slack/HTTP, HMAC-SHA256)
- 🌐 External Integrations – GitHub API (PR automation), Slack Webhooks (notifications), Custom Webhooks (HTTP POST), Audio Notifications (marisa_owattaze.wav)
- 🤖 LLM Providers – OpenAI (GPT-5-codex), Google Gemini (2.5 Pro/Flash), Anthropic (Claude 3.5+), Local/Ollama (Llama models)
# Interactive TUI
codex
# Start with prompt
codex "explain this codebase"
# Non-interactive execution
codex exec "add logging to API endpoints"
# Resume previous session
codex resume --last
# Delegate to sub-agent
codex delegate code-reviewer --scope ./src
# Deep research
codex research "React Server Components best practices" --depth 3
# Blueprint execution
codex blueprint execute ./workflows/auth-flow.jsonSee AVAILABLE_COMMANDS_v1.0.0.md for complete command reference.
Main Commands:
codex- Interactive TUIcodex exec- Non-interactive executioncodex resume- Resume previous sessioncodex apply- Apply latest diff
Agent Commands:
codex delegate- Delegate to sub-agentcodex delegate-parallel- Parallel delegationcodex pair- Pair programming with supervisorcodex agent-create- Create custom agent
Blueprint Commands:
codex blueprint create- Create new blueprintcodex blueprint execute- Execute blueprintcodex blueprint list- List blueprintscodex blueprint status- Check blueprint status
Research Commands:
codex research- Deep research with citationscodex ask- Ask with @mention integration
# Windows
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-windows-x64.exe -o codex.exe
# macOS (Intel)
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-darwin-x64 -o codex
chmod +x codex
# macOS (Apple Silicon)
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-darwin-arm64 -o codex
chmod +x codex
# Linux
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-linux-x64 -o codex
chmod +x codex# Clone repository
git clone https://github.com/zapabob/codex.git
cd codex
# Build and install
cd codex-rs
cargo install --path cli --force
# Verify installation
codex --version
# codex-cli 1.0.0Create ~/.codex/config.toml:
# Codex v1.0.0 Configuration
model = "gpt-5-codex"
[model_providers.openai]
base_url = "https://api.openai.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
[sandbox]
default_mode = "read-only"
[approval]
policy = "on-request"
[blueprint]
default_mode = "orchestrated"
enable_budget_enforcement = true
enable_policy_enforcement = true
[telemetry]
enabled = true
privacy_mode = true # SHA-256 hashing for data protection
[webhooks]
github_enabled = true
slack_enabled = falseApache-2.0 - See LICENSE for details.
リリース日: 2025年11月2日
コードネーム: "Spectrum"
🎯 Blueprint Mode (Phase 1) - 完全な階層的プランニングシステム
- 実行戦略(Single/Orchestrated/Competition)
- コスト見積もり付き予算管理
- チェックポイント/再開可能な状態永続化
- セキュリティのためのポリシー強制
🌍 多言語/review - 母国語でコードレビュー
- 8言語対応: 日本語、英語、中国語、韓国語、フランス語、ドイツ語、スペイン語、ポルトガル語
- AGENTS.md統合による自動言語検出
- 各言語向けにローカライズされたシステムプロンプト
📊 Monitoring & Telemetry - プライバシー保護型分析
- データ保護のためのSHA-256ハッシング
- 構造化ログによるイベント追跡
- Webhook統合(GitHub/Slack/HTTP、HMAC-SHA256)
🤖 マルチLLM対応 - 全主要プロバイダー向け統一インターフェース
- OpenAI GPT-5-codex(プライマリ)
- Google Gemini 2.5 Pro/Flash(検索グラウンディング)
- Anthropic Claude 3.5+(代替)
- Local/Ollamaモデル(プライバシー重視)
| 機能 | 説明 |
|---|---|
| 🤖 自動オーケストレーション | タスクの複雑さを自動分析し、専門サブエージェントを並列調整 |
| ⚡ 2.6倍高速 | 独立したタスクの並列実行で最大生産性 |
| 🔍 Deep Research | 引用付きマルチソース研究、矛盾検出(Gemini CLI + DuckDuckGo) |
| 🔒 デフォルトで安全 | 承認ポリシー付きサンドボックス分離(Seatbelt/Landlock) |
| 🌐 マルチIDE対応 | VS Code、Cursor、Windsurf拡張機能で統一体験 |
| 🔌 MCP統合 | 双方向Model Context Protocolサポート(クライアント&サーバー) |
| 🌍 クロスプラットフォーム | ネイティブパフォーマンスでWindows、macOS、Linuxをサポート |
v1.0.0アーキテクチャは10の主要レイヤーと55以上のコアコンポーネントで構成されています。
詳細は上記の英語セクションのMermaid図を参照してください。
# インタラクティブTUI
codex
# プロンプトで開始
codex "このコードベースを説明して"
# 非インタラクティブ実行
codex exec "APIエンドポイントにロギングを追加"
# 前のセッションを再開
codex resume --last
# サブエージェントに委譲
codex delegate code-reviewer --scope ./src
# Deep Research
codex research "React Server Components ベストプラクティス" --depth 3
# Blueprint実行
codex blueprint execute ./workflows/auth-flow.json完全なコマンドリファレンスは AVAILABLE_COMMANDS_v1.0.0.md を参照してください。
主要コマンド:
codex- インタラクティブTUIcodex exec- 非インタラクティブ実行codex resume- 前のセッションを再開codex apply- 最新の差分を適用
エージェントコマンド:
codex delegate- サブエージェントに委譲codex delegate-parallel- 並列委譲codex pair- スーパーバイザーとペアプログラミングcodex agent-create- カスタムエージェント作成
Blueprintコマンド:
codex blueprint create- 新規Blueprint作成codex blueprint execute- Blueprint実行codex blueprint list- Blueprint一覧codex blueprint status- Blueprint状態確認
リサーチコマンド:
codex research- 引用付きDeep Researchcodex ask- @メンション統合で質問
# Windows
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-windows-x64.exe -o codex.exe
# macOS (Intel)
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-darwin-x64 -o codex
chmod +x codex
# macOS (Apple Silicon)
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-darwin-arm64 -o codex
chmod +x codex
# Linux
curl -L https://github.com/zapabob/codex/releases/download/v1.0.0/codex-linux-x64 -o codex
chmod +x codex# リポジトリをクローン
git clone https://github.com/zapabob/codex.git
cd codex
# ビルドとインストール
cd codex-rs
cargo install --path cli --force
# インストール確認
codex --version
# codex-cli 1.0.0~/.codex/config.toml を作成:
# Codex v1.0.0 設定
model = "gpt-5-codex"
[model_providers.openai]
base_url = "https://api.openai.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
[sandbox]
default_mode = "read-only"
[approval]
policy = "on-request"
[blueprint]
default_mode = "orchestrated"
enable_budget_enforcement = true
enable_policy_enforcement = true
[telemetry]
enabled = true
privacy_mode = true # データ保護のためのSHA-256ハッシング
[webhooks]
github_enabled = true
slack_enabled = falseApache-2.0 - 詳細は LICENSE を参照してください。