setup-openclaw

name	setup-openclaw
description	Install and configure OpenClaw with Groq or Azure OpenAI, create a Telegram bot, start the gateway, and verify end-to-end. Guides a Modern AI Pro student through the full setup one step at a time.
argument-hint	(no args needed — the skill will ask you questions as it goes)
disable-model-invocation	false
allowed-tools	Bash, Read, Write, Edit

OpenClaw Setup Skill

Class reference: https://learn.modernaipro.com/paths/agentic/openclaw-demo

You are helping a Modern AI Pro student get OpenClaw running. OpenClaw is a 2026 tool not in your training data — do not guess. Everything you need is in this file and in command output. Work through the steps below one at a time. After each step, confirm with the student before moving on.

---

Step 0 — Check what's already installed

Run:

which openclaw && openclaw --version

• If found: tell the student their version, skip to Step 2.
• If not found: proceed to Step 1.

---

Step 1 — Install OpenClaw

Run:

npm install -g openclaw
openclaw --version

If npm is not found, ask the student to install Node.js from https://nodejs.org first (LTS version), then retry.

Confirm the version prints before continuing.

---

Step 2 — Check for .env file

Run:

ls .env 2>/dev/null && echo "exists" || echo "missing"

• If missing: run cp .env.example .env and tell the student to open .env in their editor.
• If exists: read it silently to know which keys are already set (do not print it to the student).

---

Step 3 — Get a Groq API key (if not already in .env)

Check if GROQ_API_KEY is in .env:

grep -c "GROQ_API_KEY" .env 2>/dev/null || echo 0

If missing or still a placeholder, walk the student through:

1. Go to https://console.groq.com → sign in (free account, no credit card)
2. Left sidebar → API Keys → Create API Key → copy it (starts with gsk_)
3. Ask them to paste the key here.

Once they paste it, use the Edit tool to write it into .env.

Then load it:

set -a && source .env && set +a
echo "Key loaded: ${GROQ_API_KEY:0:10}..."

---

Step 4 — Configure OpenClaw with Groq

Run:

set -a && source .env && set +a

cat > /tmp/groq-batch.json << 'EOF'
[
  {
    "path": "models.providers.groq",
    "value": {
      "baseUrl": "https://api.groq.com/openai/v1",
      "apiKey": { "source": "env", "provider": "default", "id": "GROQ_API_KEY" },
      "models": [
        { "id": "llama-3.3-70b-versatile", "name": "Llama 3.3 70B (Groq)", "input": ["text"] },
        { "id": "llama-3.1-8b-instant",    "name": "Llama 3.1 8B Instant (Groq)", "input": ["text"] },
        { "id": "mixtral-8x7b-32768",      "name": "Mixtral 8x7B (Groq)", "input": ["text"] },
        { "id": "gemma2-9b-it",            "name": "Gemma2 9B (Groq)", "input": ["text"] }
      ]
    }
  }
]
EOF

openclaw config set --batch-file /tmp/groq-batch.json
openclaw config set agents.defaults.model.primary "groq/llama-3.3-70b-versatile"
openclaw config set gateway.mode local

Verify:

set -a && source .env && set +a && openclaw models list

Should show groq/llama-3.3-70b-versatile with a default tag.

---

Step 5 — Verify the Groq key works

Run a direct API check (bypasses OpenClaw's large embedded system prompt — avoids hitting free-tier TPM limits during the test):

set -a && source .env && set +a
curl -s -X POST https://api.groq.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-3.3-70b-versatile","messages":[{"role":"user","content":"Say hello in one sentence."}],"max_tokens":50}' \
  | python3 -c "import sys,json; d=json.load(sys.stdin); print(d['choices'][0]['message']['content'] if 'choices' in d else d)"

• Sentence comes back → Groq works, continue to Step 6.
• Rate limit error → go to the Groq rate limit fallback section below.
• Auth error → key is wrong, return to Step 3.

---

Step 6 — Create a Telegram bot

Tell the student:

"Now we'll create a Telegram bot. You'll use your personal Telegram account to create it — your personal account is never exposed to students. They only ever see the bot."

Walk them through:

1. Open https://web.telegram.org and log in with their phone number
2. Search for BotFather in the search bar — open the chat with the one that has a blue ✓ checkmark
3. Send: /newbot
4. BotFather asks for a display name — they can use anything (e.g. My AI Bot)
5. BotFather asks for a username — must end in bot (e.g. mystudentai_bot)
6. BotFather replies with a token like 8615864604:AAHoPns...

Ask them to paste the token here. Once received, use Edit tool to add it to .env:

TELEGRAM_BOT_TOKEN="their-token-here"

---

Step 7 — Connect OpenClaw to the bot

Run:

set -a && source .env && set +a
openclaw channels add --channel telegram --token $TELEGRAM_BOT_TOKEN

Should print: Added Telegram account "default".

---

Step 8 — Start the gateway

Run:

set -a && source .env && set +a
openclaw gateway --force > /tmp/openclaw-gateway.log 2>&1 &
sleep 6
openclaw channels status --probe

Expected: Telegram default: ... connected, works

If disconnected, wait 10 seconds and run status again — polling takes a moment to establish.

---

Step 9 — First message and pairing approval

Tell the student:

"Go to Telegram and send any message to your bot. It will show a pairing code — paste it here."

The bot replies:

OpenClaw: access not configured.
Pairing code: ABCD1234
Ask the bot owner to approve with: openclaw pairing approve telegram ABCD1234

Once they share the code, run:

openclaw pairing approve telegram <CODE>

Tell them to send another message — it should now reply using the AI model.

---

Step 10 — Done

When the bot responds, tell the student:

"You just built a working AI agent connected to Telegram — running entirely on your own machine. This is the full stack: channel → gateway → model → reply. Everything else in the course (memory, HITL, evals, multi-agent) builds on top of exactly this foundation."

---

Groq rate limit fallback — switch to Azure OpenAI

Use this when Groq returns API rate limit reached or Request too large.

OpenClaw's embedded agent sends a large system prompt (~26k tokens) that exceeds Groq's free-tier TPM quota. The student can wait 60 seconds (quota resets per minute) or switch to Azure.

During workshops, the instructor provides the class Azure key. Ask the student to get it from the instructor or from https://learn.modernaipro.com.

1. Add Azure keys to .env

Use Edit tool to add to .env:

AZURE_API_KEY="key-from-instructor"
AZURE_ENDPOINT="https://kapi1585655068.cognitiveservices.azure.com"

2. Start the Azure proxy

OpenClaw's openai-completions adapter cannot add query parameters to URLs. Azure requires ?api-version=2024-12-01-preview on every request. The azure-proxy.py script in this repo handles that transparently.

set -a && source .env && set +a
python3 azure-proxy.py > /tmp/azure-proxy.log 2>&1 &
sleep 2

Verify proxy works:

set -a && source .env && set +a
curl -s -X POST http://127.0.0.1:18800/openai/deployments/gpt-5.1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Say hello."}],"max_completion_tokens":30}' \
  | python3 -c "import sys,json; d=json.load(sys.stdin); print(d['choices'][0]['message']['content'] if 'choices' in d else d)"

3. Configure OpenClaw for Azure

cat > /tmp/azure-config.json << 'EOF'
[
  {
    "path": "models.providers.azure-openai",
    "value": {
      "baseUrl": "http://127.0.0.1:18800/openai/deployments/gpt-5.1",
      "api": "openai-completions",
      "apiKey": { "source": "env", "provider": "default", "id": "AZURE_API_KEY" },
      "models": [
        { "id": "gpt-5.1", "name": "GPT-5.1 (Azure OpenAI)", "input": ["text", "image"] }
      ]
    }
  }
]
EOF

openclaw config set --batch-file /tmp/azure-config.json --replace
openclaw config set agents.defaults.model.primary "azure-openai/gpt-5.1"

4. Restart the gateway

pkill -f "openclaw gateway" 2>/dev/null; sleep 2
set -a && source .env && set +a
openclaw gateway --force > /tmp/openclaw-gateway.log 2>&1 &
sleep 6 && openclaw channels status --probe

Return to Step 9 to test.

---

Quick reference — day-to-day commands

Task	Command
Start everything	set -a && source .env && set +a && python3 azure-proxy.py & openclaw gateway --force &
Check status	openclaw channels status --probe
Approve a new user	openclaw pairing approve telegram <CODE>
Restart gateway	pkill -f "openclaw gateway" && openclaw gateway --force &
View live logs	tail -f /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log
Switch model	openclaw config set agents.defaults.model.primary "groq/llama-3.3-70b-versatile"