Why I built TekmerDB

Rafa

I'm Rafa, a database and infrastructure engineer with over a decade of experience keeping production systems — PostgreSQL, MySQL, MongoDB, TimescaleDB — reliable at scale: high-availability clusters, replication, monitoring, the works. That kind of work trains one instinct above all others: never trust a system's output until you know exactly how confident you should be in it, and never let two conflicting signals get quietly resolved without someone noticing.

In more recent projects working with AI agents and the tooling around them, I kept running into the opposite instinct. Agents are handed "memory," but it's usually just a flat store of text — no sense of confidence, no provenance, no way to know when two things it "remembers" actually contradict each other. The agent just picks one and moves on, silently.

That bothered me the same way an unmonitored production database would. So I built TekmerDB: a memory layer for AI agents that scores confidence mechanically — not guessed by an LLM — and treats disagreement between sources as a signal to surface, not a mess to hide. It runs fully air-gapped, and it's built to hold up in regulated environments where "trust me" isn't good enough, starting with the energy sector under the EU AI Act.

Get in touch

TekmerDB isn't a business — it's an idea I wanted to put out into the world to help anyone get more reliable, more honest output from their agentic AI systems. It's open source, and I'd genuinely like to hear from people using it, breaking it, or with ideas for where it should go next.