All work
Sambung
A WhatsApp-first inbox where an AI agent answers routine questions and hands the conversation to your team when a person is needed.
- Category
- AI operations
- Stage
- In development
- Built with
- Python, FastAPI, LangGraph, OpenRouter, PostgreSQL, React, Tailwind CSS, Langfuse, MCP
Demo data
About the project
The challenge
An AI assistant is only useful when the team behind it can see what it said, step in, and keep its knowledge current. Messaging channels add their own operational work.
What we built
Sambung combines a WhatsApp inbox, an operator view with human takeover, and a knowledge workbench the business owns. Agents run on LangGraph and reach models through OpenRouter, so the model can change without rebuilding the product.
Scope
- Supervised conversations with takeover, return and resolve
- Owned knowledge base and agent workbench
- Model usage and cost traced per conversation
Stack
- Python
- FastAPI
- LangGraph
- OpenRouter
- PostgreSQL
- React
- Tailwind CSS
- Langfuse
- MCP
- Kimi
- Gemini
- GPT
- Claude
Models are reached through OpenRouter and can be changed per task.
Demo data
Demo data
Demo data
Demo data
Next project: MyFC
A CRM for financing casework: intake, document checks, consultation and bank submission, tracked case by case across companies.
Demo data
