AI Agent Framework 比較

chatweb.ai vs OpenClaw vs Picoclaw vs Claude Code vs ChatGPT — 音声対応・マルチチャネル・サーバーレスRustランタイム

📊 Stats 📋 Comparison ✨ Key Features 🛠️ All Features
35
Built-in Tools
13
Channels
7
LLM Providers
130+
API Endpoints
<50ms
Cold Start
chatweb.ai
Rust (axum) • AWS Lambda
MIT 13 Channels Voice-First
OpenClaw
TypeScript • Node.js
203K
GitHub Stars
MIT 13+ Channels
Picoclaw
Go • Single Binary
13.9K
GitHub Stars
MIT Lightweight
Claude Code
TypeScript • Anthropic
36K
GitHub Stars
Apache 2.0 Coding Agent
ChatGPT
Proprietary • OpenAI
Proprietary Multi-modal Largest User Base

📋 詳細比較

Feature CHATWEB.AI OPENCLAW PICOCLAW CLAUDE CODE CHATGPT
Core
Language Rust (axum) TypeScript (Node.js) Go TypeScript Proprietary (Python)
Deployment AWS Lambda (serverless) Local / VPS / Docker Single binary (local) Local / SaaS SaaS only
Cold Start <50 ms ~1000 ms (Node.js) <1s (Go) N/A N/A (cloud)
License MIT MIT MIT Apache 2.0 Proprietary
Self-hostable ✓ Lambda / Docker / local ✓ ✓ ✗ ✗
Channels
Channel Count 13 (Web, LINE, Telegram, Discord, Slack, Teams, Facebook, Zalo, Google Chat, Matrix, Signal, iMessage) 13+ (WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Teams...) 5+ (Telegram, Discord, QQ, DingTalk, LINE) 3 (Terminal, IDE, GitHub) 4 (Web, iOS, Android, Desktop)
Voice (STT + TTS) ✓ Push-to-talk, auto-TTS ○ Wake + Talk mode ✗ ✗ ✓ Advanced Voice Mode
Cross-Channel Sync ✓ /link command + QR code ✓ ✗ ✗ ✗
AI Models
Supported Models Claude, GPT-4o, Gemini, Groq, DeepSeek, Qwen, Kimi Claude, GPT (via API keys) OpenAI, Anthropic, Gemini, Ollama, vLLM Claude only GPT-4o / GPT-5.2
Auto Failover ✓ Parallel provider racing + fallback ✗ ✗ ✗ ✗
Bring Your Own Key ✓ ✓ Required ✓ Required ✗ ✗
Load Balancing ✓ Round-robin + failover ✗ ✗ ✗ ✗ (internal)
Local LLM ✓ candle + Qwen3 GGUF ✗ ✓ Ollama / vLLM ✗ ✗
Memory
Memory System ✓ 2-layer (daily log + long-term) Session + transcript File-based (AGENTS.md, SOUL.md) CLAUDE.md + conversation ✓ Cross-conversation
Auto Consolidation ✓ Every 10 entries → LLM summarizes ✗ ✗ ✗ ✗
Context Summarization ✓ Older messages auto-compressed ✗ ✗ ✓ ✓
Tools & Agent
Built-in Tools 35 (24 core + 11 optional; web search, code exec, file ops, image/music/video gen, browser...) 50+ (ClawHub skills) File ops, workspace tools fs, git, shell, MCP Code Interpreter, DALL-E, Web Browse
Sandboxed Code Exec ✓ Shell/Python/Node.js sandbox ✓ Docker sandbox ○ Workspace-restricted ✓ ✓ Python sandbox
Agentic Loop 1-5 iterations (plan-based) Continuous Interactive CLI 7 parallel agents Continuous (GPT-5.2)
MCP Support ✓ JSON-RPC 2.0 ✗ ✗ ✓ ✗
Multi-modal ○ Text + Voice ○ Text + Voice Text only Text only ✓ Text + Image + Voice + Video
API & Developer
REST API ✓ 130+ endpoints ○ Local Gateway ✗ ✗ ✓ OpenAI API
SSE Streaming ✓ ✓ WebSocket ✗ N/A ✓
Built-in Billing ✓ Stripe integration ✗ ✗ Anthropic billing OpenAI billing
Pricing
Pricing Credit-based (100 free, from $9/mo) Free (BYOK) Free (BYOK) $20-$200/mo Free / $8-$200/mo
Best For Voice-first multi-channel AI Privacy-first personal assistant Minimal self-hosted agent Developer coding workflows General-purpose AI assistant

✨ 主な差別化ポイント

🎤 Voice-First

唯一のオープンソースAIエージェントフレームワークで、ネイティブSTT + TTS + Push-to-talk UIを搭載。電話のように話しかけるだけ。音声入力時は応答を自動で音声再生。

web/index.html — Web Speech API + /api/v1/speech/synthesize

📱 13 Channels

LINE, Telegram, Discord, Slack, Teams, Facebook, Zalo, Google Chat, Matrix, Signal, iMessage等。/linkコマンドとQRコードでクロスチャネル会話同期。OpenClawは13+、Picoclawは5+、ChatGPTは4チャネル。

crates/nanobot-core/src/channel/ — line.rs, telegram.rs, facebook.rs, ...

⚡ Fastest Runtime

Rust on Lambda ARM64 = <50 ms cold start、<2s応答。Picoclaw (Go)は<1s、OpenClaw (Node.js)は~1000 ms。~24MB binary (ARM64)。

crates/nanobot-lambda/ — Compiled Rust on AWS Lambda

🔄 Auto Failover

プライマリモデルが失敗？全プロバイダーを並列レースで自動フォールバック。ユーザーにはエラーゼロ。複数のAPIキーでロードバランシング。他のフレームワークにはこの機能なし。

crates/nanobot-core/src/provider/mod.rs — LoadBalancedProvider

🧠 Long-Term Memory

2層メモリシステム：デイリー会話ログ + 長期記憶。10エントリーごとに安価なLLMで自動統合。Picoclawは SOUL.md ファイル、ChatGPT はクロス会話メモリがあるが、どちらも自動統合なし。

crates/nanobot-core/src/service/http.rs — read_memory_context(), append_daily_memory()

☁️ Serverless Scale

AWS Lambdaでゼロから無限にスケール。VPS管理不要。DynamoDB single-table design。OpenClawとPicoclawはセルフホスティング必要。ChatGPTはクラウドのみでセルフホスト不可。

infra/template.yaml — SAM template for Lambda + DynamoDB + API Gateway

🛠️ 全機能一覧

🧠 AI Models & Providers

  • OpenAI GPT-4o / GPT-4o-mini
  • Anthropic Claude Sonnet / Opus
  • Google Gemini 2.0 Flash / Pro
  • Groq (Llama, Mixtral)
  • DeepSeek Chat / Reasoner (R1)
  • Qwen, Kimi (Moonshot)
  • Local LLM fallback (candle + Qwen3 GGUF)
  • Multiple API keys with round-robin load balancing
  • Auto failover (primary → gpt-4o-mini → gemini-flash)

🛠 35 Built-in Tools (24 core + 11 optional)

  • Web Search — Brave/Bing/Jina 3-tier fallback
  • Web Fetch — Jina Reader for JS-heavy pages
  • Browser — CSS selector, screenshots, form submit
  • Code Execution — Sandboxed shell/Python/Node.js
  • File Read/Write/List — Per-session sandbox
  • Filesystem — find (glob), grep
  • Image Generate — OpenAI gpt-image-1
  • Music Generate — Suno API
  • Video Generate — Kling API
  • Calculator / Weather / Wikipedia / Translation
  • YouTube Transcript / arXiv Search / News Search
  • QR Code / DateTime / CSV Analysis
  • Google Calendar / Gmail (OAuth) — API key required
  • Slack / Discord / Notion / PostgreSQL / Spotify — API key required
  • Phone Call (Amazon Connect) / Webhook Trigger — API key required
  • Web Deploy — S3/CloudFront subdomain — API key required
  • GitHub — Read/Write files, Create PRs — API key required

📱 13 Channels

  • Web — Voice-first SPA with push-to-talk
  • LINE — Webhook + auto-link via QR
  • Telegram — Deep link + /start
  • Facebook Messenger
  • Discord, Slack, Microsoft Teams
  • Zalo, Google Chat
  • Signal, Matrix, iMessage (BlueBubbles)
  • REST API & MCP for AI agents

🧠 Memory System

  • Daily conversation logs (DAILY#{date})
  • Long-term memory (LONG_TERM)
  • Auto-consolidation every 10 entries
  • Yesterday's context for continuity
  • Context summarization for long conversations
  • Memory API (GET/DELETE /api/v1/memory)
  • View & clear memory from Settings UI

🔧 Developer Platform

  • REST API — 130+ endpoints
  • SSE streaming responses
  • MCP server (JSON-RPC 2.0)
  • API Playground with shareable results
  • API key management
  • Slash commands (/link, /share, /help, /status)
  • Settings & Memory APIs
  • Webhook support for all channels

⚙️ Infrastructure

  • Rust on AWS Lambda ARM64 (<50ms cold start)
  • DynamoDB single-table design
  • ~24MB binary (ARM64, single binary)
  • Stripe billing integration
  • Google OAuth authentication
  • AI agent discovery (robots.txt, llms.txt, ai-plugin.json)
  • Self-hostable (Docker, Lambda, local)
  • MIT License — 100% open source
無料で試してみる

GitHub · API Docs · Pricing