Multi-tenant AI sales agent framework powered by LangGraph
One codebase → infinite businesses → any messaging channel
A production-ready, open-source framework for building AI sales agents that serve multiple businesses from a single deployment. Each business (tenant) gets their own branded agent personality, product catalog, and LLM config — all wired up to Telegram, Instagram, WhatsApp, and your web app simultaneously.
Think of it as SalesGPT but with proper multi-tenancy, channel adapters, and a repository pattern for clean data access.
Customer on Telegram → Flower Shop's "Aigul" (GPT-4o-mini, ru) → searches catalog → sends product images
Customer on WhatsApp → Restaurant's "Marco" (Claude Sonnet, en) → shows menu + promotions
Customer on Web → E-commerce "Alex" (Llama 3, en) → visual product search
This isn't a tutorial project. I've been building AI sales agents for real businesses — flower shops, restaurants, e-commerce stores. Every client starts with this exact codebase, and I add integrations on top.
The pattern is always the same:
- Drop a
config.yamlwith the business persona and rules - Load the product catalog into
products.json - Connect the channel (usually Telegram first, then Instagram/WhatsApp)
- Go live in a day
This repo is the starting point I keep coming back to — so I'm open-sourcing it.
Need to deploy this for your business? Reach out → github.com/yerdaulet-damir or open an issue describing your use case.
The core is a shared LangGraph graph that drives every tenant's agent. Tenant YAML configs inject the personality, tools, and LLM at runtime — no code changes needed to add a new business.
┌─────────────────┐
│ Tenant Config │ ← config.yaml + products.json
│ (YAML + JSON) │ per business in /tenants/
└────────┬────────┘
│
┌──────────────────▼──────────────────┐
│ LangGraph Sales Graph │
│ │
│ START → [Assistant] ⇄ [Tools] → END│
│ │
└──────────────────────────────────────┘
↑ ↑
┌─────────────┴──────┐ ┌────────────┴──────────────┐
│ Channel Adapters │ │ Tool Registry │
│ • Telegram Bot │ │ • search_products │
│ • Instagram DM │ │ • get_promotions │
│ • WhatsApp │ │ • search_by_image 🖼️ │
│ • Web / FastAPI │ │ • send_product_image │
└────────────────────┘ └───────────────────────────┘
| Feature | This Project | SalesGPT | TaskingAI | AutoGPT |
|---|---|---|---|---|
| Multi-tenant YAML configs | ✅ Drop a folder, done | ❌ | ✅ (complex SaaS) | ❌ |
| LangGraph native (async, streaming) | ✅ | ❌ custom | ❌ | ❌ |
| Channel adapters (TG/IG/WA/Web) | ✅ Built-in | ❌ | ❌ | |
| Image search / visual catalog | ✅ | ❌ | ❌ | ❌ |
| Switch LLM per tenant (OpenAI/Anthropic) | ✅ | ❌ | ✅ | ❌ |
| LangSmith tracing out of the box | ✅ | ❌ | ❌ | ❌ |
pip install + YAML = live agent |
✅ | ❌ | ❌ | ❌ |
| Open source, no SaaS lock-in | ✅ | ✅ | ❌ | ✅ |
| Stars ⭐ | growing 🌱 | ~1.2k | ~5.1k | ~177k |
The pitch: AutoGPT is powerful but general. SalesGPT is focused but doesn't scale to multiple businesses. TaskingAI is SaaS. This project solves exactly the multi-tenant sales bot problem with ~500 lines of clean Python.
enterprise-agents/sales/
├── src/
│ ├── channels/ # Channel adapters
│ │ ├── telegram/ # Telegram Bot polling
│ │ ├── instagram/ # Instagram DM webhook
│ │ ├── whatsapp/ # WhatsApp Business API
│ │ └── web/ # FastAPI + WebSocket
│ ├── graphs/
│ │ └── sales_graph.py # ⭐ Core LangGraph graph
│ ├── nodes/
│ │ ├── assistant.py # LLM node (builds prompt, routes)
│ │ └── tool_executor.py # Tool execution node
│ ├── tools/ # LangChain tools
│ │ ├── search_products.py
│ │ ├── get_promotions.py
│ │ ├── search_by_image.py
│ │ └── send_product_image.py
│ ├── repositories/
│ │ └── base.py # Repository Protocol (swap storage easily)
│ ├── config/
│ │ ├── settings.py # Global env-var settings (Pydantic)
│ │ ├── tenant_config.py # YAML tenant loader
│ │ └── llm_provider.py # OpenAI / Anthropic factory
│ ├── models/
│ │ └── product.py # Product datamodel
│ └── state/
│ └── agent_state.py # LangGraph state schema
├── tenants/
│ └── flower_shop/ # Example tenant
│ ├── config.yaml # Agent name, LLM, language, rules
│ └── products.json # Product catalog
├── main.py # CLI chat interface for testing
├── langgraph.json # LangGraph Studio config
└── .env.example # Copy to .env and fill in your keys
git clone https://github.com/yerdaulet-damir/langgraph-sales-agent
cd enterprise-sales-agent
python -m venv .venv && source .venv/bin/activate
pip install -e .cp .env.example .env
# Edit .env and add your API keys:
# ANTHROPIC_API_KEY=sk-ant-...
# OPENAI_API_KEY=sk-...
# LANGCHAIN_API_KEY=ls__... (optional, for LangSmith tracing)python main.pyYou'll see a tenant picker, then start chatting with the flower shop's AI consultant:
🏪 Available tenants:
1. flower_shop — Розы & Букеты (agent: Айгуль)
You: Show me red roses
Айгуль: Let me search for red roses for you... 🌹
[calls search_products → returns catalog results]
Айгуль: Here are our beautiful red roses! We have...
langgraph devOpen Studio UI at http://localhost:8124 — full visual graph debugger with state inspection.
- Create a folder in
/tenants/your_business/ - Add
config.yaml:
tenant_id: "your_business"
business_name: "Your Store Name"
language: "en"
agent:
name: "Alex"
role: "Sales Assistant"
personality: |
You are Alex, a friendly assistant at Your Store.
You help customers find the perfect products.
rules:
- Always greet warmly
- Suggest upsells when appropriate
- Never confirm payment yourself
llm:
provider: "anthropic" # or "openai"
model: "claude-3-5-sonnet-20241022"
temperature: 0.7
max_tokens: 1024
features:
image_search: true
promotions: true
upsell: true- Add
products.jsonwith your catalog — that's it. No code changes.
| Channel | Status | Notes |
|---|---|---|
| CLI (testing) | ✅ Ready | python main.py |
| Web / FastAPI | ✅ Ready | REST + WebSocket |
| Telegram Bot | ✅ Ready | Long polling + webhooks |
| Instagram DM | ✅ Ready | Messenger API webhook |
| WhatsApp Business | ✅ Ready | Cloud API webhook |
# At least one LLM provider key is required
OPENAI_API_KEY=your-key
ANTHROPIC_API_KEY=your-key
# LangSmith (optional but highly recommended)
LANGCHAIN_TRACING_V2=true
LANGCHAIN_API_KEY=your-langsmith-key
LANGCHAIN_PROJECT=sales-agent| Key | Required | Default | Description |
|---|---|---|---|
tenant_id |
✅ | — | Unique ID matching folder name |
business_name |
✅ | — | Shown in system prompt |
language |
✅ | — | en, ru, kk, etc. |
agent.name |
✅ | — | Agent persona name |
agent.role |
✅ | — | Role description |
agent.personality |
✅ | — | Free-form prompt injection |
agent.rules |
✅ | — | List of hard rules |
llm.provider |
✅ | — | openai or anthropic |
llm.model |
✅ | — | Model ID |
features.image_search |
❌ | false | Enable visual product search |
features.promotions |
❌ | false | Enable promotions tool |
| Layer | Technology |
|---|---|
| Agent framework | LangGraph 0.4+ |
| LLM providers | OpenAI, Anthropic (swappable per tenant) |
| Config | Pydantic v2 + PyYAML |
| Web server | FastAPI + Uvicorn |
| Tracing | LangSmith |
| Python | 3.11+ |
- PostgreSQL / Redis product repository (swap JSON backend)
- Voice channel (Telegram voice messages → STT → agent)
- Stripe payment link generation tool
- Order tracking tool
- Multi-language auto-detection
- Docker Compose deployment guide
- More tenant examples (restaurant, e-commerce)
PRs welcome! This project is intentionally small and readable — the core graph is ~50 lines. Good first issues:
- Add a new channel adapter (Discord, Viber, Line)
- Add a new tool (order status, payment link)
- Add a new tenant example
- Improve the repository pattern (PostgreSQL backend)
# Development setup
pip install -e ".[dev]"
ruff check src/ # Linting
pytest # Tests- ✅ No secrets in git —
.envis gitignored - ✅ Tenant isolation — each tenant only sees its own products
- ✅ No hardcoded keys — all config via environment variables (
python-dotenv+pydantic) ⚠️ InMemorySaver — conversation state is in-memory only. For production, swap withPostgresSaverorRedisSaver.
MIT — use it, fork it, sell it, build your empire.
Built with ❤️ using LangGraph · Inspired by the need for a simple, hackable multi-tenant sales bot framework
If this helped you ship faster, star ⭐ the repo