{
  "item": "OpenClaw",
  "fields": {
    "project_name": "OpenClaw",
    "type": "Personal AI assistant / multi-channel AI gateway with extensible messaging integrations",
    "current_status": "Active development — frequent releases (versioned YYYY.M.D, e.g. 2026.4.15). Production-ready stable channel; beta and dev channels also available. Rapidly growing community. Open-source with commercial-friendly MIT license. Sponsored by OpenAI, GitHub, NVIDIA, Vercel, Blacksmith, Convex.",
    "license": "MIT",
    "repository": "https://github.com/openclaw/openclaw",
    "website": "https://openclaw.ai",
    "docs": "https://docs.openclaw.ai",
    "skills_registry": "https://clawhub.com",

    "architecture": {
      "runtime_model": "Node.js 24+ (Node 22.16+ minimum). Single long-lived Gateway process (daemon) that owns all messaging surfaces. Gateway exposes a typed WebSocket API (JSON frames) on a configurable bind host (default 127.0.0.1:18789). Control-plane clients (CLI, macOS app, web UI, automations) connect over WebSocket. Nodes (macOS/iOS/Android/headless) also connect over WebSocket with role=node. One Gateway per host. Protocol: first frame must be 'connect', then req/res + server-push events. TypeBox schemas define the protocol; JSON Schema generated from schemas; Swift models generated for iOS/macOS. Supports launchd (macOS) and systemd (Linux) daemon supervision.",
      "channel_support": "WhatsApp (via Baileys), Telegram (via grammY), Slack, Discord, Google Chat, Signal, iMessage (BlueBubbles), iMessage legacy, IRC, Microsoft Teams, Matrix (E2EE supported), Feishu/Lark, LINE, Mattermost, Nextcloud Talk, Nostr, Synology Chat, Tlon, Twitch, Zalo, Zalo Personal, WeChat, QQ, WebChat (built-in web UI), macOS (native), iOS/Android (via node apps). 20+ channels total.",
      "persistence": "File-based workspace at ~/.openclaw/workspace. Session transcripts stored in sessions.json. Memory in MEMORY.md (long-term) and memory/YYYY-MM-DD.md (daily). Optional LanceDB integration for vector memory search (supports cloud storage since 2026.4.x). Config in ~/.openclaw/openclaw.json. DM pairing stored in local allowlist store. Cron job manifest in manifest.db.",
      "extensibility": "Skills system (SKILL.md-based plugin/workflow files). Bundled skills included. Managed skills via ClawHub registry. Workspace skills (~/.openclaw/workspace/skills/). Skills can be community-contributed. Plugin system via bundled channel lazy-loaded plugins (partitioned by OPENCLAW_BUNDLED_PLUGINS_DIR). MCP registry integration referenced. Multi-agent routing supported — route channels/accounts/peers to isolated agent workspaces. Docker sandboxing for non-main sessions."
    },

    "features": {
      "agent_capabilities": "Full agentic loop with tool use. Browser control (Playwright/Chromium). Canvas (live visual workspace with A2UI). File read/write/edit. Shell exec (with PTY support). Web fetch. Image generation, video generation, music generation, TTS (ElevenLabs + system TTS + Google Gemini TTS). PDF/DOCX/PPTX/XLSX handling. Memory search (semantic, LanceDB). Subagent spawning and session management. Cron jobs. Webhooks. Heartbeat (proactive check-ins). Voice Wake + Talk Mode (macOS/iOS/Android). Camera and screen capture (Android node). Code execution via Claude Code / Codex / Cursor agent delegation.",
      "skills_plugin_system": "Skills are SKILL.md + optional supporting scripts/files. Three sources: bundled (shipped with openclaw), managed (installed via CLI), workspace (user-created, in workspace/skills/). Skill registry: ClawHub (clawhub.com). Skills auto-injected into agent system prompt as available_skills. Each skill can have its own tools, instructions, and dependencies. Agent can self-modify and create new skills via conversation. Skills snapshots cached per session with invalidation on config changes.",
      "webhook_api": "Webhook inbound supported (docs at docs.openclaw.ai/automation/webhook). Gmail Pub/Sub integration for push email. WebSocket-based Gateway RPC API accessible to external clients over WS (with token auth). REST-like CLI wrapper. Remote access via Tailscale or SSH tunnel. Typed WebSocket protocol with request/response + server-push event model.",
      "cron_scheduling": "Built-in cron job system. Jobs in ~/.openclaw/cron/jobs.json. Managed via 'openclaw cron' CLI. Supports isolated cron sessions (separate from main session). NO_REPLY silent sentinel support. Cron announce delivery with media/caption handling. Jobs can run in any connected channel.",
      "message_logging": "Full session transcript persistence in sessions.json. Daily memory logs in memory/YYYY-MM-DD.md. Logging docs at docs.openclaw.ai/logging. Gateway debug logging available (--verbose flag). WebChat + macOS app include debug tools. CLI: /trace on|off, /verbose on|off, /usage tokens|full. Control UI with Model Auth status card and exec approval modal."
    },

    "operational": {
      "setup_complexity": "Low to moderate. Guided CLI onboarding: 'openclaw onboard --install-daemon' walks through gateway, workspace, channels, and skills setup step by step. Works on macOS, Linux, Windows (WSL2). Daemon auto-installs as launchd/systemd service. Channel setup requires creating bots/API keys (e.g., BotFather for Telegram, Discord app). DM pairing flow for access control. 'openclaw doctor' for diagnostics. Community reports 5-30 minutes to first working setup.",
      "documentation_quality": "High quality. Comprehensive docs at docs.openclaw.ai covering: getting started, onboarding wizard, channel guides, concepts/architecture, tools/skills, cron/webhooks, security, remote access, Docker, troubleshooting. DeepWiki integration (deepwiki.com/openclaw/openclaw). FAQ. llms.txt index for AI-assisted doc traversal. Actively maintained alongside releases.",
      "self_hosting": "Fully self-hosted. Gateway runs on any Linux/macOS/Windows(WSL2) host. No cloud dependency required. Models from any provider (Anthropic, OpenAI, Google, local via Ollama/LM Studio, etc.). Optional: Docker sandboxing for multi-user/channel safety. Remote access via Tailscale or SSH tunnel. Nix flake available (github.com/openclaw/nix-openclaw). Docker image available.",
      "resource_usage": "Node.js process (typically 100-300MB RAM for gateway). Additional per LLM API calls (external providers, no local inference required by default). LanceDB for vector memory is optional (adds disk I/O). Multiple concurrent channel connections (WhatsApp Baileys is persistent WS, Telegram via grammY polling). Low CPU at idle; spikes during agent turns. Can run on VPS, Raspberry Pi, Mac Mini.",
      "update_maintenance": "npm-based updates: 'npm install -g openclaw@latest' or 'openclaw update --channel stable|beta|dev'. 'openclaw doctor' checks config/health. Automatic daemon restart via launchd/systemd. CLI prunes stale dist chunks after upgrades. Changelog maintained at CHANGELOG.md. Active release cadence (multiple releases per week during growth phase). GitHub CI/CD."
    },

    "best_for": [
      "Personal AI assistant that operates 24/7 across all messaging channels you already use",
      "Power users who want a self-hosted, hackable, privacy-respecting AI assistant",
      "Developers who want to automate workflows via conversation (build skills, run code, manage files)",
      "Multi-channel inbox management (WhatsApp + Telegram + Discord + Slack + email in one agent)",
      "Proactive autonomous tasks: cron jobs, heartbeats, background monitoring, reminders",
      "Teams wanting a shared AI assistant on their own infrastructure",
      "Integration-heavy use cases: webhooks, Gmail Pub/Sub, GitHub, calendar, APIs",
      "Voice-first workflows (macOS/iOS/Android wake words + TTS)",
      "Users who want skills/plugins that self-improve (agent can create new skills via chat)"
    ],

    "weaknesses": [
      "Single-user focus by design — multi-user setups require Docker sandboxing and careful config",
      "Windows support only via WSL2 (no native Windows gateway)",
      "Requires LLM API keys (Anthropic Claude recommended) — adds ongoing cost beyond server cost",
      "Learning curve for channel setup (each channel requires its own bot/API key creation)",
      "WhatsApp integration uses unofficial Baileys library — risk of account ban if abused",
      "No built-in billing/quota management — must monitor LLM spend externally",
      "Young project (launched ~early 2026) — rapid breaking changes between releases possible",
      "Memory/dreaming system complex to tune — multiple storage modes, LanceDB optional",
      "Remote access requires SSH tunnel or Tailscale (no built-in secure public endpoint)"
    ],

    "same_vps_coexistence": {
      "summary": "Excellent coexistence profile. OpenClaw runs as a single Node.js process (gateway) on a configurable port (default 18789, loopback only). It does not require root, does not bind public ports by default, and uses minimal system resources at idle. Can share a VPS with Hermes or other AI assistant frameworks without port conflicts. Uses systemd user service (not system service). Docker sandboxing option available for process isolation. No database server required (file-based by default). Tailscale or SSH tunnel for remote access — no public port exposure needed.",
      "port": "18789 (default, loopback 127.0.0.1 only)",
      "process": "Single openclaw/openclaw.mjs Node.js process + optional companion apps",
      "daemon": "systemd user service or launchd (macOS)",
      "conflicts": "None expected with standard VPS workloads. Check port 18789 is free."
    }
  },

  "sources": [
    "https://openclaw.ai (marketing/testimonials page, fetched 2026-04-18)",
    "https://github.com/openclaw/openclaw (GitHub README, fetched 2026-04-18)",
    "https://docs.openclaw.ai/concepts/architecture (architecture docs, fetched 2026-04-18)",
    "/usr/lib/node_modules/openclaw/README.md (local installed package)",
    "/usr/lib/node_modules/openclaw/package.json (local package manifest, version 2026.4.15)",
    "/usr/lib/node_modules/openclaw/CHANGELOG.md (local changelog, entries for 2026.4.15)"
  ],

  "uncertain": [
    "Exact RAM/CPU numbers at idle — estimated 100-300MB based on typical Node.js usage; not benchmarked",
    "WhatsApp ban risk specifics — Baileys is unofficial; risk level depends on usage patterns",
    "ClawHub (clawhub.com) skill registry details — site not fetched; based on README reference only",
    "Hermes framework details — this research file covers OpenClaw only; Hermes comparison requires separate research",
    "LanceDB cloud storage provider specifics — mentioned in changelog but details not fetched from docs",
    "Exact GitHub star count / community size — referenced as growing but not quantified here",
    "iOS/Android companion app availability in app stores vs sideload only — not confirmed from fetched sources"
  ]
}