B2B SDR Agent Template

We're live on Product Hunt! → PulseAgent B2B SDR Agent Skill — Open-source AI SDR for WhatsApp, Email & Telegram. An upvote means the world to us 🙏

Turn any B2B export business into an AI-powered sales machine in 5 minutes.

✅ New · 2026-06-07 — Template v3.8.0: GitHub Actions CI for npm test, post-deploy deploy/doctor.sh health checks, configurable remote OpenClaw paths for non-root deployments, and expanded validation for skill counts + workspace path generation. See full changelog →

🚀 New · 2026-06-09 — OpenClaw v2026.6.5: Parallel web-search integration for richer prospect research, ClawHub skill installation with pinned commits for safer deployments, Anthropic extended-thinking session recovery, Google Chat native approval cards, WhatsApp config-reload teardown fix, SQLite-backed auth profiles, and QQBot reasoning-tag filtering for cleaner channel replies. See full changelog →

📦 Add-on · 2026-05-22 — WhatsApp Legacy Account Onboarding Spec v0.5 (Path D — Multi-Device fetch): Lowest-friction history acquisition — customer scans one QR, waits for Multi-Device sync, bootstrap.sh pulls .txt files straight from PulseAgent. No backup, no password, no USB. Backup-extraction (v0.4) still default for zero-downtime tenants. Three-layer architecture (MemOS + sales_playbook + conversation_history), A/B/C/D delivery decision tree, strict auto-onboard gate. Release notes →

An open-source, production-ready template for building AI Sales Development Representatives (SDRs) that handle the full sales pipeline — from lead capture to deal closing — across WhatsApp, Telegram, and email.

Built on OpenClaw, battle-tested with real B2B export companies.

🌐 English | 中文 | Español | Français | العربية | Português | 日本語 | Русский

Architecture: 7-Layer Context System

┌─────────────────────────────────────────────────┐
│              AI SDR Agent                       │
├─────────────────────────────────────────────────┤
│  IDENTITY.md   → Who am I? Company, role        │
│  SOUL.md       → Personality, values, rules     │
│  AGENTS.md     → Full sales workflow (10 stages)│
│  USER.md       → Owner profile, ICP, scoring    │
│  HEARTBEAT.md  → 14-item pipeline inspection    │
│  MEMORY.md     → 4-layer anti-amnesia protocol  │
│  TOOLS.md      → CRM, channels, integrations    │
├─────────────────────────────────────────────────┤
│  Skills        → Extensible capabilities        │
│  Product KB    → Your product catalog           │
│  Cron Jobs     → 14 automated pipeline checks   │
├─────────────────────────────────────────────────┤
│  OpenClaw Gateway (WhatsApp / Telegram / Email) │
└─────────────────────────────────────────────────┘

Each layer is a Markdown file you customize for your business. The AI reads all layers on every conversation, giving it deep context about your company, products, and sales strategy.

Quick Start

Option A: OpenClaw Users (1 Command)

If you already have OpenClaw running:

clawhub install b2b-sdr-agent

Browse more skills on ClawHub — the AI agent skill marketplace.

Done. The skill installs the full 7-layer context system, delivery-queue, and sdr-humanizer into your workspace. Then customize:

# Edit the key files for your business
vim ~/.openclaw/workspace/skills/b2b-sdr-agent/references/IDENTITY.md
vim ~/.openclaw/workspace/skills/b2b-sdr-agent/references/USER.md

# Or copy to your main workspace
cp ~/.openclaw/workspace/skills/b2b-sdr-agent/references/*.md ~/.openclaw/workspace/

Replace all {{placeholders}} with your actual company info, and your AI SDR is live.

Option A2: Hermes Agent Users (1 Command)

If you use Hermes Agent instead of OpenClaw:

hermes skills install github:iPythoning/b2b-sdr-hermes-skill

Same 10-stage SDR pipeline, adapted for Hermes conventions (native cron, skill config, built-in WhatsApp). Migrating from OpenClaw? Run hermes claw migrate first, then install the skill.

→ b2b-sdr-hermes-skill repo

Option B: Full Deployment (5 Minutes)

1. Clone & Configure

git clone https://github.com/iPythoning/b2b-sdr-agent-template.git
cd b2b-sdr-agent-template

# Edit the 7 workspace files for your business
vim workspace/IDENTITY.md   # Company info, role, pipeline
vim workspace/USER.md       # Your products, ICP, competitors
vim workspace/SOUL.md       # AI personality and rules

2. Set Up Deployment Config

cd deploy
cp config.sh.example config.sh
vim config.sh               # Fill in: server IP, API key, WhatsApp number

3. Validate Locally

cd ..
npm test

This checks the template structure, shell scripts, product KB JSON, proforma invoice generation, Chroma memory smoke flow, and generated OpenClaw config before you touch a server.

4. Deploy

cd deploy
./deploy.sh my-company

# Output:
# ✅ Deploy Complete: my-company
# Gateway:  ws://your-server:18789
# WhatsApp: Enabled
# ChromaDB Memory: Enabled (chromadb + local chroma-memory)
# Skills:   b2b_trade (41 skills)

5. Verify Remote Health

./doctor.sh

Doctor checks SSH, Node/npm, OpenClaw CLI, openclaw.json permissions, workspace context files, local skills, Gateway service state, Gateway /health, and Chroma memory.

That's it. Your AI SDR is live on WhatsApp and ready to sell.

What It Does

Full-Pipeline Sales Automation (10 Stages)

Automated Scheduling (14 Pipeline Checks)

Human-Like Conversations

• Responds in the customer's language (auto-detected)
• Natural message pacing with delivery-queue skill (3-90s delays, message splitting)
• Timezone-aware sending (7 market timezone mappings)
• Cultural adaptation (Middle East, Africa, SEA, LatAm, Europe)
• Never reveals AI identity — presents as your sales consultant

Dynamic ICP Scoring

• Initial score based on 5 weighted dimensions (purchase volume, product match, region, payment ability, authority)
• Auto-adjusts based on interaction: fast reply +1, asks for quote +2, mentions competitor +2, 7d no reply -1
• Hot leads (ICP>=7) auto-flagged, owner notified immediately

4-Layer Anti-Amnesia System

AI agents lose context in long conversations and across sessions. Our 4-layer anti-amnesia architecture ensures your AI SDR never forgets:

Message In ──→ L1 MemOS auto-recall (structured memory injection)
    │
    ├──→ L3 ChromaDB per-turn store (customer-isolated, auto-tagged)
    │
    ├──→ L2 Proactive Summary at 65% tokens (haiku compression, zero info loss)
    │
    └──→ L4 CRM Snapshot daily 12:00 (disaster recovery fallback)

Result: Your AI SDR remembers every customer, every quote, every commitment — even after 100+ turns, weeks of silence, or system restarts.

See ANTI-AMNESIA.md for the full implementation spec with code, prompts, and deployment guide.

The 7 Layers Explained

Skills

Pre-built capabilities that extend your AI SDR:

Skill Profiles

Choose a pre-configured skill set based on your needs:

Industry Examples

Ready-to-use configurations for common B2B export verticals:

To use an example, copy it into your workspace:

cp examples/heavy-vehicles/IDENTITY.md workspace/IDENTITY.md
cp examples/heavy-vehicles/USER.md workspace/USER.md
# Then customize for your specific business

Product Knowledge Base

Structure your product catalog so the AI can generate accurate quotes:

product-kb/
├── catalog.json                    # Product catalog with specs, MOQ, lead times
├── products/
│   └── example-product/info.json   # Detailed product info
└── scripts/
    └── generate-pi.js              # Proforma invoice generator

Control Dashboard

After deployment, your AI SDR comes with a built-in web dashboard:

http://YOUR_SERVER_IP:18789/?token=YOUR_GATEWAY_TOKEN

The dashboard shows:

• Real-time bot status and WhatsApp connection
• Message history and conversation threads
• Cron job execution status
• Channel health monitoring

The token is auto-generated during deployment and printed in the output. Keep it private — anyone with the URL+token has full access.

Security note: Set GATEWAY_BIND="loopback" in config.sh to disable remote dashboard access. Default is "lan" (accessible from network).

WhatsApp Configuration

By default, the AI SDR accepts messages from all WhatsApp contacts (dmPolicy: "open"). This is the recommended setting for sales agents — you want every potential customer to be able to reach you.

To change after deployment, edit ~/.openclaw/openclaw.json on the server:

{
  "channels": {
    "whatsapp": {
      "dmPolicy": "open",
      "allowFrom": ["*"]
    }
  }
}

Then restart: systemctl --user restart openclaw-gateway

Deployment

Prerequisites

• A Linux server (Ubuntu 20.04+ recommended)
• Node.js 18+
• An AI model API key (OpenAI, Anthropic, Google, Kimi, etc.)
• WhatsApp Business account (optional but recommended)

Configuration

All configuration lives in deploy/config.sh. Key sections:

# Server
SERVER_HOST="your-server-ip"

# AI Model
PRIMARY_API_KEY="sk-..."

# Channels
WHATSAPP_ENABLED=true
TELEGRAM_BOT_TOKEN="..."

# CRM
SHEETS_SPREADSHEET_ID="your-google-sheets-id"

# Admin (who can manage the AI)
ADMIN_PHONES="+1234567890"

For non-root servers or custom OpenClaw locations, set the remote paths instead of editing deploy scripts:

SERVER_USER="openclaw"
REMOTE_HOME="/srv/openclaw"
REMOTE_OPENCLAW_HOME="/srv/openclaw/.openclaw"
REMOTE_CONFIG_HOME="/srv/openclaw/.config"
REMOTE_WORKSPACE_DIR="/srv/openclaw/.openclaw/workspace"

CI runs npm test on every push and pull request to main.

WhatsApp IP Isolation (Multi-Tenant)

When running multiple agents on the same server, each should have a unique exit IP so WhatsApp sees independent devices. This prevents cross-account flagging.

# After deploying a client, isolate their WhatsApp IP:
./deploy/ip-isolate.sh acme-corp

# Or with a specific SOCKS5 port:
./deploy/ip-isolate.sh acme-corp 40010

How it works:

                  ┌─ wireproxy :40001 → WARP Account A → CF IP-A
                  │    ↑
tenant-a ─────────┘    ALL_PROXY=socks5://host:40001

tenant-b ─────────┐    ALL_PROXY=socks5://host:40002
                  │    ↓
                  └─ wireproxy :40002 → WARP Account B → CF IP-B

Each tenant gets:

• A dedicated free Cloudflare WARP account
• An isolated wireproxy instance (~4MB RAM)
• A unique Cloudflare exit IP for all outbound traffic (including WhatsApp)

To auto-enable during deploy, set IP_ISOLATE=true in config.sh.

Managed Deployment

Don't want to self-host? PulseAgent offers fully managed B2B SDR agents with:

• One-click deployment
• Dashboard & analytics
• Multi-channel management
• Priority support

Get Started →

Community

Join the conversation, ask questions, and share your use cases:

• Discord — Real-time chat with the team
• Telegram — Community discussions
• X / Twitter — Announcements & updates
• Product Hunt — Upvote & reviews

Learn more:

• PulseAgent Blog — Tutorials, release notes, and case studies
• Open Source Hub — All repos, docs, and ecosystem links
• ClawHub — Browse and install AI agent skills
• Running on Kimi K2.6 — LLM setup guide (pitfalls, auth profiles, context budgeting)

Contributing

Contributions welcome! See our Contributing Guide for details. Areas where we'd love help:

• Industry templates: Add examples for your industry
• Skills: Build new capabilities
• Translations: Translate workspace templates to other languages
• Documentation: Improve guides and tutorials

License

MIT — use it for anything.

Built with ❤️ by PulseAgent
Context as a Service — AI SDR for B2B Export