Rick Houlihan Builds mneme to Bridge AI and Oracle Gaps

Rick Houlihan, a veteran with thirty years of experience shipping enterprise systems, teamed up with Claude to build and write code directly in the trenches. He highlights that the most critical engineering challenges lie in the gap between what official documentation claims and how systems actually behave under production workloads.

During a work session last month, Houlihan spent hours chasing a non-existent table after a CreateTable call returned a CREATING status with zero errors. Even after provisioning a second database in a different service tier to rule out capacity and permission limits, the API returned the exact same misleading response. While the Oracle ADB Developer tier documentation suggested the feature was fully supported, the system in practice failed to deliver.

This discrepancy between documentation and runtime reality costs engineering teams weeks of troubleshooting, yet it represents invaluable institutional knowledge that no manual has ever documented.

To address this, the team originally built an AI plugin for Claude Code and Codex named oracle-ai-dev by ingesting SQL references and developer guides. However, they realized that an AI trained solely on manuals is akin to a well-read new hire who has never handled a midnight system outage. The real solution required a loop that automatically translates hands-on troubleshooting into reusable plugin updates.

Named mneme, this closed-loop engine passively captures operational knowledge during real work, stages it locally for human review, and submits a pull request to an open Git repository. A comprehensive field survey conducted on 2026-08-11 across more than forty systems revealed a distinct gap in the market: no existing tool combined automated capture with open Git-based team federation. This unmatched demand also aligns with an open feature request anthropics/claude-code#38536 filed on 2026-03-25.

Source: Dev.to