agentic-data-stack
Ask BigQuery or Snowflake questions in plain English — no BI subscription required
Overview
Point an LLM at your data warehouse and ask questions in plain English. The agent writes the SQL, runs it, and returns results.
ClickHouse built the original stack around two open-source projects they acquired: LibreChat (chat interface) and Langfuse (LLM observability). This fork extends it with BigQuery and Snowflake support.
What’s Included
| Service | Role |
|---|---|
| LibreChat | Chat UI with multi-model support (Claude, GPT, Gemini) |
| Langfuse | Observability — trace every query, token, and tool call |
| MCP Toolbox for Databases | Exposes warehouse schema and SQL execution as MCP tools |
Quick Start
# 1. Generate secrets and env files
./prepare-demo.sh
# 2. Configure your warehouse connection in tools.yaml
# (BigQuery, Snowflake, or ClickHouse)
# 3. Launch the stack
docker compose up
