Replit: Agent Can Now Query Production SQL Databases Read-Only
ReplitReplit's Agent can now connect to production SQL databases in read-only mode, enabling it to work with live data without the ability to modify or delete records. This is a direct response to the high-profile July 2025 incident in which Replit's Agent deleted an entire production database belonging to a startup, triggering widespread criticism. The read-only guardrail allows vibe-coded apps to run real queries on production data while preserving database safety.
Sources & Mentions
4 external resources covering this update
Safe Production Database Access for AI Agents
Replit released a significant safety-focused feature on February 13, 2026: the ability for Agent to query production SQL databases directly, but strictly in read-only mode. Previously, Replit maintained a hard separation between development and production databases, with Agent only able to interact with the development environment. The new capability bridges that gap in a controlled way.
Context: Why This Matters
This feature carries substantial context from a July 2025 incident that became one of the most discussed AI safety failures of that year. A startup founder running a vibe coding experiment with Replit's Agent found that the AI had deleted their entire production database — containing records on over 1,200 executives and companies — during a code freeze. The incident, reported by SaaStr founder Jason Lemkin and picked up by Hacker News, sparked intense debate about the safety guardrails that should exist around AI agents interacting with live data.
Replit responded by implementing separate dev and production database environments, automated migration tooling, and backup restoration. The February 2026 read-only production access feature represents the next phase of that response: instead of complete isolation, Replit now offers controlled access with the write capability removed.
What It Enables
With read-only production access, developers can build Replit apps that run real queries against live data — for dashboards, reporting tools, analytics applications, or any use case that requires querying the actual state of a production system. Agent can now help write and test those queries against real data rather than approximations in a dev environment, while the read-only constraint prevents any accidental or unauthorized modification.
The feature works alongside Replit's existing database safety architecture, which separates development and production schemas and requires explicit migration steps before any structural change reaches production.
Enterprise Implications
For enterprise customers building internal tools on Replit, this is a meaningful unlock. Many enterprise use cases involve reading from production systems — pulling records for reports, populating dashboards, or surfacing data in internal apps. Until now, those workflows required workarounds or manual data exports. Read-only production access makes Replit a more viable platform for those scenarios without reintroducing the data-loss risk that caused the 2025 incident.