apM Claw

Your AI intern working from AM to PM.
Protecting crypto communities the apM way, powered by the apM Claw engine.

Heritage

Built on OpenClaw, Rebuilt for Crypto Security

apM Claw is an independent project inspired by OpenClaw's memory-context architecture (MIT License). However, less than 10% of the original codebase remains—this is a ground-up rebuild focused on Telegram spam defense.

What Changed:

• OpenClaw: General-purpose agent framework (Slack, Discord, Web UI, Telegram)
• apM Claw: Telegram-only security appliance for crypto communities
• Removed: 80% of features (cron, browser, sessions, subagents, image generation, multi-platform support, Docker)
• Added: Spam forwarding, crypto-specific threat detection, code-level admin verification

PM-E's Journey:

• Since 2023: PM-E served as apM community's guardian
• 2025: Rebuilt from scratch as a specialized security tool

Security Architecture

Design Philosophy: "Lean & Strong Claw"

PM-E is a purpose-built security appliance for Telegram crypto communities. Our core principle:

LLMs cannot be trusted with authorization decisions.

Real protection happens in TypeScript code and Telegram API calls, not system prompts.

---

Key Security Measures

1. Attack Surface Reduction

• 6 tools only: message, spam_delete, memory_search, memory_get, web_search, web_fetch
• Removed: browser, subagents, sessions_*, cron, image, session_status, memory_save, Docker support

2. Code-Level Authorization

// Admin verification via Telegram API (not prompts)
const member = await telegram.getChatMember(chatId, userId);
const isAdmin = ["administrator", "creator"].includes(member.status);

3. Spam Handling

• Certain spam → spam_delete (forward to archive + delete)
• Uncertain spam → ignored silently (admin handles manually)

4. Group Access Control

• groupPolicy: "allowlist" — only responds in registered groups
• Spam archive chat middleware filter — prevents re-detection loop of forwarded spam
• Pre-startup message filter — ignores messages queued during downtime

5. Network Security

• HTTPS-only, local network blocking (127.0.0.0/8, 10.0.0.0/8, 192.168.0.0/16)

---

Current Moderation Features

Implemented:

• Silent spam deletion via spam_delete (certain spam)
• Role-based admin exemption (code-level, not prompt-level)
• Spam archive forwarding before deletion
• Startup chat title + admin status logging

In Development:

• User-level sanctions (ban/mute for repeat offenders)
• Surge detection with auto "Slow Mode"

---

Threat Model

PM-E defends crypto communities against:

• Coordinated spam waves (50+ accounts)
• Phishing campaigns (wallet drainers, airdrop scams)
• Impersonation attacks (Unicode tricks)
• Pump-and-dump shilling

---

Deployment

Environment

pnpm install
pnpm build

Configuration

Managed via GitHub Secrets (.env + config/apmclaw.json auto-generated on deploy):

Secret	Purpose
TELEGRAM_TOKEN_*	Bot token (separate for dev/prod)
TELEGRAM_CHAT_IDS_*	Monitored group IDs (comma-separated)
TELEGRAM_CHAT_IDS_FORWARD_SPAM_*	Spam archive chat ID
AWS_BEARER_TOKEN_BEDROCK	Bedrock API auth
BRAVE_SEARCH_API_KEY	Web search

Run

# Run with PM2
pm2 start dist/entry.mjs --name apmclaw -- gateway --bind lan --allow-unconfigured --port 18790

Auto-deployed via GitHub Actions (deploy-dev.yml, deploy-prod.yml):

• dev: push to dev branch → port 18789
• prod: push to main branch → port 18790

Workspace Templates

PM-E's behavior is defined by templates in docs/reference/templates/:

• SOUL.md - Identity and communication style
• AGENTS.md - Operation guidelines and spam patterns
• TOOLS.md - Available capabilities
• MEMORY.md - apM project wiki (links, news, exchanges)

These are auto-generated to workspace/ on first run.

Contribution

As an evolving project, PM-E grows through community interaction and open-source contributions. Check our repository: https://github.com/apmcoin/apmclaw

---

Made with the apM way
Protect the community with intelligence, humility, and proactive silence.