Building an Agentic Chatbot with Weave

Goal

Using our own products regularly helps us better empathize with and understand our users' needs. This repo provides a streamlined guide to experience how Weave works in a typical AI development workflow.

Go from zero to a production-deployed support bot with systematic evaluation, real-time monitoring, and continuous improvement.

Project

Build a support bot for Weights & Biases that can:

• Answer questions about our product (from our docs)
• Create and give updates on support tickets

Your Task

Get this bot ready for production. Going from 0 to demo is easy, but can you build an agent ready to face real customer questions? Discover:

• Where Weave shines in the development process
• What features are intuitive vs. confusing
• What's missing or could be improved

Prerequisites

• Python 3.12+ environment
• GitHub to clone the repo
• Terminal access to run commands
• Modal account (free) for serverless deployment (sign up here)
  • Used to deploy your agent server (both development and production)
  • Free tier includes generous compute credits
• Weights & Biases account (sign up free)
  • Your W&B API key is used for both Weave observability AND the LLM API (we use W&B Inference with DeepSeek)

---

Getting Started 🚀

1. Clone the repository

git clone https://github.com/wandb/agentic-support-bot-demo.git
cd agentic-support-bot-demo

2. Install dependencies

We use uv for fast, reliable Python package management.

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install project dependencies (includes Marimo)
uv sync

3. Launch the Interactive Guide

uv run marimo run marimo-guide.py

The guide will open in your browser and walk you through the entire tutorial.

---

What You'll Learn

The interactive guide covers:

1. Project Setup - Configure environment variables and workspace
2. Basic Agent - Get a minimal agent running with Weave tracing
3. Tools & MCP - Add capabilities and connect to documentation search
4. Iteration - The core Weave workflow: observe → diagnose → fix → verify
5. Evaluation - Build systematic evaluation with datasets and scorers
6. Deployment - Deploy to Modal for production
7. Monitoring & Guardrails - Add safety controls and quality tracking

---

Resources

• Slide Framework Documentation
• Tyler CLI Documentation
• Weave Documentation
• Creating Custom Tools Guide
• Get W&B API Key

---

License

See LICENSE file for details.