The Reviews That Changed How MindsDB Ships
BACKGROUND
MindsDB builds open-source AI infrastructure for the enterprise.
MindsDB is the AI data platform behind Anton, an autonomous BI agent that lets teams query their operational data in plain language. Founded in Berkeley in 2017 and backed by
Benchmark, Mayfield, NVIDIA, and Y Combinator, MindsDB powers tens of thousands of developers and enterprise teams.
Their codebase spans 191 repositories. A query engine connecting to 200+ data sources, an autonomous agent running LLM-driven workflows, and an enterprise platform managing credentials across customer environments. The surface area for subtle bugs is wide.
THE PROBLEM
Speed at the cost of safety nets.
Every PR can introduce something subtle. A SQL string that should have been parameterized. A try/except that swallows the wrong exception. An auth check that falls through on the unhappy path.
GitHub Copilot was already in their review pipeline. But the team needed a reviewer that actually read the codebase, learned from their own past mistakes, and flagged what would break in production, not what would break in a test environment.
THE SWITCH
Fourteen days. 191 repos. One bot doing real work.
MindsDB activated Entelligence on April 16, 2026. Over two weeks, the bot reviewed 78 PRs across 21 active repos, left 142 inline comments, and ran 103 review executions with zero failures. Average execution time: 76 seconds. No CI drag. No flaky retries. Fast, focused review on every non-draft PR.
RESULTS
43% of comments changed the code.
Of the 142 inline comments Entelligence left, 61 were verified findings. The developer either resolved the review thread or changed the flagged code in a subsequent commit. That is a 43% hit rate across 26 distinct PRs.
WHAT WAS CAUGHT
Security bugs that would have shipped.
Five distinct security issues were flagged and fixed before merge. Real code, on production-bound branches, on a platform that handles enterprise customer data.
Correctness bugs caught before merge.
Eleven runtime-crash-class bugs were flagged and fixed in the same window. NameErrors, message-ordering bugs, memory-write races, and a React Router upgrade that silently broke navigation state.
ASK ELLIE
The whole codebase, in your chat.
Beyond automated review, MindsDB's engineering team adopted Ask Ellie, Entelligence's conversational interface that gives engineers instant access to cross-repo context without leaving their workflow.
"Ask Ellie changed how I debug and investigate issues. I get the full engineering context in chat, what changed, where it changed, and what else might be affected. Before, I was hopping between four different dashboards to triangulate anything. Now I just ask." Lucas Koontz, Principal Engineer, MindsDB |
For a team shipping across 21 active repos, the ability to query engineering history, trace change impact, and surface relevant past incidents in a single conversation is real time saved per engineer, per day, without requiring a new tool, a new tab, or a new workflow.
THE LEARNING LOOP
Cross-repo memory is shipping work.
On auth #66, Entelligence flagged a top-level import of hubspot that would break the graceful install flow for optional dependencies. The bot cited a guideline learned from a prior anton review: optional dependencies must be imported locally, not at the top level.
RELIABILITY
Seventy-six seconds. Zero failures.
Across 103 review executions during the trial window, Entelligence had zero failures. The bot ran on every non-draft PR, kept pace with a team shipping across 21 active repos, and never blocked CI.
Hit rate is the only metric that matters.
A code review tool that nobody acts on is just noise. MindsDB's engineers acted on 43% of Entelligence's comments and 0% of Copilot's in the same window. Five security bugs and eleven correctness bugs were caught and fixed before merge. The cross-repo learning is real. The reliability is operational.
Faster shipping. Fewer incidents. A team that trusts every PR.
Try Entelligence now!