LangGraph sales agent
A personal open-source sandbox for sales agents, with 45 GitHub stars. It answers customer questions using a business’s product catalog. My commercial AI Sales Manager SaaS for WhatsApp and Instagram is a separate product.
What it does
The repository includes a flower-shop example: the agent looks up products and uses the catalog to answer questions. You can try it in a command-line chat.
Each business has its own catalog, instructions, and model settings.
GitHub repository
Read the code, installation instructions, example configuration, and MIT license.
How it works
The shared LangGraph workflow selects tools and works with the current tenant’s catalog. YAML configuration and a JSON product catalog define a business without duplicating the graph. The repository supports model configuration for OpenAI and Anthropic.
This is useful when several businesses need the same sales workflow with different products and instructions. Channel adapters and business integrations are separate deployment work.
Running it
The public starter uses in-memory conversation storage. A production deployment needs durable storage such as PostgreSQL or Redis, appropriate authentication, and isolation checks for its adapters.
The project targets Telegram, Instagram, WhatsApp, and web. Channel adapters need implementation and verification for each deployment.
Related work
I also created iCombooster, an AI sales CRM for enriched, ICP-scored lead profiles, and worked on B2B CRM and automation products at Latenode.
iCombooster
25,000 leads enriched with 40+ attributes.
Latenode
Product discovery, UX, implementation, and rollout of business automation.