A beautiful web client for OpenClaw.

curl -fsSL https://opencami.xyz/install.sh | bash

npm install -g opencami

opencami --gateway ws://127.0.0.1:18789 --token <GATEWAY_TOKEN>

Then open: http://localhost:3000

This is the safest setup: Gateway stays on loopback, you access OpenCami via https://<magicdns>:<port>.

1. In OpenClaw, allowlist the exact OpenCami URL (no trailing slash):

{
  "gateway": {
    "controlUI": {
      "allowedOrigins": ["https://<magicdns>:3001"]
    }
  }
}

2. Restart the gateway:

openclaw gateway restart

3. Start OpenCami with the same origin:

opencami \
  --gateway ws://127.0.0.1:18789 \
  --token <GATEWAY_TOKEN> \
  --origin https://<magicdns>:3001

⚠️ Note: --gateway must be ws:// or wss:// (not https://).

opencami [--port <n>] [--host <addr>] [--gateway <ws(s)://...>] [--token <token>] [--password <pw>] [--origin <url>] [--no-open]

--port <n>        Port to listen on (default: 3000)
--host <addr>     Host to bind to (default: 127.0.0.1)
--gateway <url>   OpenClaw gateway WS URL (default: ws://127.0.0.1:18789)
--token <token>   Gateway token (sets CLAWDBOT_GATEWAY_TOKEN)
--password <pw>   Gateway password (sets CLAWDBOT_GATEWAY_PASSWORD)
--origin <url>    Origin header for backend WS (sets OPENCAMI_ORIGIN)
--no-open         Don't open browser on start
-h, --help        Show help

You can also set env vars instead of flags:

CLAWDBOT_GATEWAY_URL=ws://127.0.0.1:18789
CLAWDBOT_GATEWAY_TOKEN=...
OPENCAMI_ORIGIN=https://<magicdns>:3001   # only needed for remote HTTPS

• "origin not allowed" → add the exact URL to gateway.controlUI.allowedOrigins and pass the same value as --origin / OPENCAMI_ORIGIN (exact match, no trailing /).
• Pairing required → approve the device in OpenClaw (openclaw devices list/approve).
• Fallback (only if needed): OPENCAMI_DEVICE_AUTH_FALLBACK=1

• Prefer wss:// for remote connections.
• Prefer token auth (CLAWDBOT_GATEWAY_TOKEN) over password.
• Keep allowedOrigins minimal (exact origins only, no wildcards).
• Treat OPENCAMI_DEVICE_AUTH_FALLBACK=true as temporary compatibility mode.
• Do not expose OpenCami directly to the public internet without TLS + access controls.
• For Tailnet deployments, limit Tailnet device/user access.

Cause: gateway rejected browser origin.

Fix:

1. Add origin to gateway.controlUI.allowedOrigins (exact match, no trailing /)
2. Set identical OPENCAMI_ORIGIN (or --origin) in OpenCami
3. Restart gateway (openclaw gateway restart)

Cause: gateway auth succeeded but the device was paired with insufficient scopes.

Fix (v1.8.5+): delete the device identity and let OpenCami re-pair automatically:

rm ~/.opencami/identity/device.json
# then restart OpenCami — it will re-pair with full scopes

On first connect, OpenCami registers itself as a device on the gateway. Starting with v1.8.5, this happens automatically with full scopes (operator.admin, operator.approvals, operator.pairing) — no manual config required.

If you see a "device pending" error:

openclaw devices list    # find the pending device
openclaw devices approve <deviceId>

After approval, OpenCami reconnects and stores a deviceToken for future sessions (no shared token needed).

Checks:

openclaw gateway status
echo "$CLAWDBOT_GATEWAY_URL"
echo "$CLAWDBOT_GATEWAY_TOKEN"

Also verify URL scheme (ws:// local, wss:// remote).

docker build -t opencami .
docker run -p 3000:3000 opencami

• ⚡ Real-time streaming — persistent WebSocket + SSE, token-by-token
• 📎 File attachments — upload PDFs, text, code, CSV, JSON via attach button or drag & drop (/uploads/ + read tool workflow)
• 📄 File cards — uploaded files render as clickable cards (filename, icon, size) and open in File Explorer
• 🖼️ Image attachments — drag & drop with compression (images stay Base64 for vision)
• 🔊 Voice playback (TTS) — ElevenLabs → OpenAI → Edge TTS fallback
• 🎤 Voice input (STT) — ElevenLabs Scribe → OpenAI Whisper → Browser
• 🔔 Browser notifications — background tab alerts when assistant replies

• 🏷️ Smart titles — LLM-generated session titles
• 💡 Smart follow-ups — contextual suggestions after each response
• 🧠 Thinking level toggle — reasoning depth (off/low/medium/high) per message
• 🔎 Search sources badge — see which search engines were used
• 📊 Context window meter — visual token usage indicator

• 📂 File explorer — browse & edit 30+ file types with built-in editor
• 🧠 Memory viewer — browse and edit MEMORY.md and daily memory files
• 🤖 Agent manager — create, edit, delete agents from the sidebar
• 🧩 Skills browser — discover and install skills from ClawHub
• ⏰ Cron jobs panel — manage scheduled automations
• 🔧 Workspace settings — toggle each tool on/off in Settings

• 🎨 Model selector — switch AI models per message
• 🎭 Persona picker — 20 AI personalities
• 🦎 Chameleon theme — light/dark/system with accent colors
• 🔤 Text size — S / M / L / XL
• 🔌 Multi-provider LLM — OpenAI, OpenRouter, Ollama, or custom

• 📁 Session folders — grouped by kind (chats, subagents, cron, other)
• 📌 Pin sessions — pinned always on top
• 🗑️ Bulk delete — select multiple sessions, delete at once
• 🛡️ Protected sessions — prevent accidental deletion
• 📥 Export — Markdown, JSON, or plain text

• 📱 PWA — installable, offline shell, auto-update
• 🖥️ Tauri desktop app (Beta) — native wrapper for macOS/Windows/Linux
• ⌨️ Keyboard shortcuts — full power-user navigation
• 💬 Slash commands — inline help and actions
• 🔍 Conversation search — current (⌘F) and global (⌘⇧F)

git clone https://github.com/robbyczgw-cla/opencami.git
cd opencami
npm install
cp .env.example .env.local
npm run dev

Then open the URL printed by Vite in your terminal.

Dev port notes: this repo's npm run dev script uses port 3002. If you run Vite directly with the config default, it targets 3003 and auto-falls back to the next free port.

Note: The desktop app is experimental and under active development. The primary focus of OpenCami is the web app. Native builds (desktop & mobile) are secondary.

OpenCami can also run as a native macOS/Windows/Linux desktop wrapper built with Tauri v2. The app loads your self-hosted OpenCami web instance.

• Node.js 18+
• Rust toolchain (rustup)

# Install dependencies (if not already done)
npm install

# Build web assets first
npm run build

# Build desktop app
npm run tauri:build

By default, the desktop app connects to http://localhost:3003.

To override at build time:

OPENCAMI_REMOTE_URL="https://your-server.example.com" npm run tauri:build

Built installers/bundles are written to src-tauri/target/release/bundle/:

• macOS: .app, .dmg
• Windows: .exe, .msi
• Linux: .deb, .AppImage

• Tray icon (hide to tray on close)
• Native notifications
• Auto-start on login
• Custom titlebar
• Multiple windows (⌘N)
• Clipboard integration

npm run tauri:dev

Requires a display/GUI environment.

• Features
• Desktop App (Tauri)
• Architecture
• Deployment
• FAQ
• Contributing
• Changelog

Built on top of WebClaw by @ibelick.

File Explorer by @balin-ar (PR #2).

Dockerfile by @deblanco (PR #7).

Powered by OpenClaw.

• 🌐 opencami.xyz
• 📦 npm
• 💻 GitHub

MIT