A safe way to let coding agents interact with your database (without prod write access)

We previously examined common approaches teams use to protect production databases (i.e. command allowlists, SQL filters, and manual approval workflows) and why they fail in the presence of autonomous agents. The primary reason is that agents "work really hard" - they often route around these restrictions to deliver the results with any possible execution surface (shell, file system, runtime). This tutorial demonstrates how to grant database access in Pochi without exposing production credentials or enabling uncontrolled writes. Why this matters Agents must never execute arbitrary code against production systems. At the same time, agents are most useful when they can read and write data to iterate quickly. The challenge is doing this safely. We’ll walk through multiple access tiers, explain their security tradeoffs, and show how to progressively increase agent autonomy without expanding the production attack surface. Prerequisites Before we start, you’ll need the following: Node.js ins