Built by a developer,
for developers

The story behind Kronvex

My name is Baptiste Hoffmann. I'm a software engineer based in Paris. I hold an engineering degree in computer science, specialised in data science, and spent three years as a data analyst at Thales Digital Identity & Security — building and maintaining systems that processed a database of 150 million patents. Working at that scale taught me what it really means when infrastructure fails silently.

At Thales, I automated patent data collection via API integrations, built Power BI dashboards for IP strategy teams, and applied machine learning to competitive intelligence workflows. I also spent four months at Safran in Cincinnati, Ohio, working on supply chain analytics — structuring ERP data, building KPI dashboards, and automating reporting pipelines in Python and SQL. In both cases: production data, real business stakes, no room for flaky tooling.

When I started building AI agents on the side, I ran into the same wall every time: the agents were stateless. Everything they learned during a session — user preferences, past decisions, accumulated context — vanished the moment the conversation ended. I explored the available memory solutions. Most required a dedicated vector database, custom embedding pipelines, and weeks of configuration. The managed options were US-hosted and opaque on pricing.

So I built the layer I wished existed. FastAPI backend, PostgreSQL with pgvector for cosine similarity, OpenAI text-embedding-3-small, hosted entirely in the EU on Supabase Frankfurt. Three endpoints: /remember, /recall, /inject-context. I launched Kronvex in March 2026. No VC funding — EU-hosted by principle, because that's the right default when you're building infrastructure that stores what AI agents know about your users.

What we stand for

Built on

The stack is intentionally boring. Proven tools, EU infrastructure, no exotic dependencies.