Deploy a Cloudflare Worker that connects Even Realities G2 smart glasses to your OpenClaw Gateway.

G2 Glasses → (voice→text) → CF Worker → OpenClaw Gateway → Full Agent
                                ↓                              ↓
                          G2 display (text)            Telegram (rich content)

• Short tasks (chat, questions): Gateway responds → displayed on G2
• Long tasks (write code, articles): G2 shows "Working on it..." → result sent to Telegram
• Image generation: DALL-E generates → sent to Telegram (G2 can't show images)
• Fallback: If Gateway is down, falls back to direct Claude API

1. Even Realities G2 glasses with Even app (v0.0.7+ with "Add Agent" support)
2. OpenClaw Gateway with HTTP API enabled
3. Cloudflare account (free plan works)
4. Anthropic API key (fallback)
5. Optional: OpenAI API key (image gen), Telegram bot token (rich content delivery)

On your OpenClaw host, enable the chat completions endpoint:

openclaw config set gateway.http.endpoints.chatCompletions.enabled true
openclaw gateway restart

Verify:

curl -X POST https://YOUR_GATEWAY_URL/v1/chat/completions \
  -H "Authorization: Bearer YOUR_GATEWAY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"openclaw","messages":[{"role":"user","content":"hi"}]}'

Copy scripts/worker.js to your project, then deploy:

# Install wrangler
npm install -g wrangler

# Login to Cloudflare
wrangler login

# Deploy
wrangler deploy

Or use the Cloudflare Dashboard: Workers & Pages → Create → Upload worker.js.

# Required
wrangler secret put GATEWAY_URL      # Your OpenClaw Gateway URL
wrangler secret put GATEWAY_TOKEN    # Your Gateway auth token
wrangler secret put G2_TOKEN         # Token for G2 glasses auth (you choose)
wrangler secret put ANTHROPIC_API_KEY # Fallback when Gateway is down

# Optional (for Telegram delivery of rich content)
wrangler secret put TELEGRAM_BOT_TOKEN
wrangler secret put TELEGRAM_CHAT_ID

# Optional (for image generation)
wrangler secret put OPENAI_API_KEY

In Even app → Settings → Add Agent:

• Name: Your agent name (e.g., "Cloud Lobster")
• URL: https://YOUR_WORKER.workers.dev
• Token: The G2_TOKEN you set above

curl -X POST https://YOUR_WORKER.workers.dev \
  -H "Authorization: Bearer YOUR_G2_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"openclaw","messages":[{"role":"user","content":"Hello, who are you?"}]}'

1. G2 converts speech → text, sends as OpenAI chat completion format
2. Worker authenticates via G2_TOKEN
3. Worker classifies request:
   • Image gen → DALL-E + Telegram (immediate G2 ack)
   • Long task → immediate G2 ack + background Gateway call → Telegram
   • Short task → proxy to Gateway → return to G2
4. Gateway runs full agent loop (memory, tools, skills)
5. Response filtered for G2 display (no URLs, code blocks → Telegram)

Two-layer token auth:

G2 --[G2_TOKEN]--> Worker --[GATEWAY_TOKEN]--> Gateway

• G2 only knows G2_TOKEN — if glasses are lost, change only this token
• GATEWAY_TOKEN stays in Worker secrets, never exposed to glasses
• Gateway HTTP API should be behind auth (token or password mode)

• 576×136 pixels, monochrome green, ~48 chars wide
• Text only (no images, no markdown rendering)
• Worker auto-filters: URLs → [link], code blocks → [code], long text → truncated
• Non-displayable content forwarded to Telegram

Edit the task classification regex in worker.js:

• isLongTask() — patterns that trigger background processing
• isImageGenRequest() — patterns that trigger DALL-E

• G2 says "Unauthorized": Check G2_TOKEN matches in Worker secrets and Even app
• "Gateway not configured": Verify GATEWAY_URL and GATEWAY_TOKEN secrets are set
• Timeout on short tasks: Gateway may be slow; increase SHORT_TIMEOUT (max ~25s for CF Workers)
• No Telegram delivery: Check TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID secrets
• Gateway returns 404: Ensure chatCompletions.enabled: true in OpenClaw config + restart