A small team with an unfashionable belief.
Models will keep changing. The hard part — giving them accurate, permission-correct context from messy company systems — will not. We work on that part.
Why we exist
Novexa Labs started in 2023 after our founders spent a year shipping internal assistants at two different companies and hit the same wall twice: the models were good enough, the data plumbing was not. Every team rebuilt the same fragile pipeline, and every one of them under-invested in access control until an incident forced the issue.
We decided to solve it as infrastructure rather than as a feature inside somebody's chat product. That means no lock-in to a single model vendor, no proprietary document format, and a product that is honest about what it does not do well.
How we work
- Engineering-led sales. The first call is with an engineer, not a rep.
- No lock-in by design. Model-agnostic, exportable indices, documented schemas.
- Ship small, measure always. Every release is gated on a public internal eval set.
Milestones
Founded in Amsterdam
Two engineers, one Postgres extension, and a very patient design partner in logistics.
First twenty production deployments
Connector library grew past forty sources. Permission mirroring shipped as a first-class feature.
Series A · EU data plane
Single-tenant deployment became the default for regulated customers. Lisbon office opened.
Four billion documents
Point-in-time retrieval and the eval harness generalised across all connectors.
Leadership
Elin Vestergaard
Co-founder & CEORafael Oliveira
Co-founder & CTOMira Kovač
Head of SecurityWe are hiring engineers who like hard data problems.
Retrieval, distributed indexing, and security engineering. Remote across CET ± 3.