NemoClaw — PolyAI Agent Studio Assistant

Containerised OpenClaw instance for PolyAI colleagues. Accessible via Slack.

What it can do

• Build new Agent Studio agents (flows, topics, functions, push to Studio)
• Analyse call data (containment rates, handoffs, transcripts) across 5 accounts
• Run the data-driven improvement loop (audit → propose → review × 2 → push)
• Answer Agent Studio questions using builder-knowledge rules

Setup

On your Mac — copy assets to EC2

# From your Mac:
EC2=ubuntu@<your-ec2-ip>
scp -r ~/.openclaw/skills/agent-studio $EC2:~/nemoclaw-studio/skills/
scp -r ~/agent-studio-builder/tools/conv-fetch $EC2:~/nemoclaw-studio/tools/
scp -r ~/nemoclaw-studio $EC2:~/

On EC2

# 1. Create .env with all API keys (see setup-aws.sh for template)
nano .env

# 2. Run setup
chmod +x setup-aws.sh
./setup-aws.sh

Slack bot setup

1. Create a new Slack app at api.slack.com
2. Add scopes: channels:history, channels:read, chat:write, groups:history, groups:read, im:history, im:read, im:write, mpim:read, reactions:write, users:read
3. Enable Socket Mode → get xapp-... token
4. Install to workspace → get xoxb-... token
5. Add both tokens to .env

Usage (in Slack)

@Aria build me a banking agent for PLATFORM account
@Aria what's the containment rate for WEX last 7 days?
@Aria audit the claim flow on wex-usp and suggest improvements
@Aria fetch me 5 transcripts from failed verifications on brightree

Architecture

Colleague → Slack → AWS EC2
                      └── Docker container
                            ├── OpenClaw gateway (Slack socket mode)
                            ├── Agent Studio skill (CLAUDE_TEMPLATE.md)
                            ├── fetch_conversations.py
                            ├── poly CLI
                            └── Claude API (Anthropic)

Maintenance

# Logs
docker compose logs -f

# Restart
docker compose restart

# Update builder-knowledge
docker compose exec polyai-studio git -C /workspace/builder-knowledge pull

# Shell into container
docker compose exec polyai-studio bash