Every idea enters through the hopper as a paper or a demo from research.bivelio.com and travels down the line until it becomes a company. You choose the station where you start.
01paper
Research
The paper enters the line.
We start from BiVelio’s open research: information retrieval, agents, graphs, orchestration and quantitative finance. You read the work, reproduce the demo and pick the real problem it solves.
Open papers and demos
Questions nobody has turned into a product yet
Direct access to the researchers
02prototype
Prototype
The demo learns to walk.
We turn the result into something people can touch: a prototype in front of real users. We iterate fast, measure what matters and drop what does not work without regret.
Prototypes on real data and cases
Short build-and-test cycles
Design from day one
03product
Build
The prototype becomes a product.
The product is built on BiVelio One, BiVelio’s operating system: identity, data, AI, the Velio assistant and agents already solved. Your team spends its time on what makes the product unique.
Identity and permissions from the start
Data and European AI built in
Velio and agents working inside
04company
Launch
The product becomes a company.
When the product stands on its own, it leaves the line as a company: with its team, its customers and BiVelio behind it. Whoever built it stays on board.
A path from product to company
Business and market support
The founding team, on board
The foundation
Everything runs on BiVelio One. Nobody starts from scratch.
Beneath the belt there is a platform: BiVelio One. Identity, data, AI, the Velio assistant and agents are already there, at every station.
Identity
Data
AI
Velio
Agents
What we put in your hands
The factory’s resources. You do not walk into an empty workshop.
This is what you find on the line from day one.
Open research
The papers and demos of research.bivelio.com, and direct conversation with their authors.
R-01
BiVelio One
BiVelio’s operating system: identity, data, AI, Velio and agents, ready for your product.
R-02
European AI
Fast, balanced and advanced models, with data handled in Europe and under its rules.
R-03
Infrastructure
Lab, test and production environments to go from demo to real use without setting anything up.
R-04
Mentors
People who have already researched, built and launched, beside you at every station on the line.
R-05
Path to company
When the product works, the support to turn it into a company: business, market and team.
R-06
Who we are looking for
Minds that want to make. What you want to build matters most.
We are looking for people who enjoy making things that work. Where you come from matters less.
Engineering
You like taking a system from the whiteboard to production, and making it hold.
Build
Research
You have written, read or reproduced papers and want to see them solve real problems.
Research
Design
You make complex things clear at first sight, on screen and in the product.
Prototype
Founders
You want to turn a technology into a company and lead the team that takes it to market.
Launch
What we share
Curiosity over titles
Show before we promise
Build in the open
Do what we truly love
Output dock
The line has shipped its first two products. The other slots are reserved.
The applications, papers and companies that leave the factory appear here. The first two are papers from BiVelio; every remaining slot is for what the people who join now will build.
01Product 01 · Paper
BiVelio Research · October 2026
BV-SALA Savings Layer: Verifiable Measurement of Cost Savings in Large Language Model Applications
How to measure savings on the bill of LLM applications with the provider’s own billing counter. Every scenario runs twice, with and without the layer; savings are split into four ledgers that are never summed, and every figure is published with its 95% confidence interval.
English, with a Spanish summary of conclusions · 7 pages
−31.5%gpt-4o-mini bill95% CI: 25.7–36.7GSM8K stream with 30% exact repeats.
−75.1%claude-sonnet-5 bill95% CI: 69.3–79.6Same stream. About 41 pp is the provider’s own prefix-cache discount, reported separately.
2.5%gateway PRO lever on Anthropic95% CI: 1.5–3.6Net saving over 51 Claude Code sessions, at list prices.
02Product 02 · Preprint
BiVelio Privacy Research Team · BiVelio · October 2026
BiVelio Shield: A Local-First, Fail-Closed Data Minimization Layer for Privacy-Preserving Use of Large Language Models
A local layer that detects and replaces sensitive data before the prompt leaves the device. Three layers: L1 deterministic, L2 credentials and L3 contextual NLP, precision-oriented, which escalates any doubt to a fail-closed gate that holds the prompt. Tiers: Free at €0 (browser only), Pro at €4 per seat per month, and Team at €8 per seat per month with a 5-seat minimum.
English, with a Spanish executive summary · 10 pages · Company technical report, not peer reviewed
1.000precision and recall on L1MOD-97 · LuhnOn checksum-validated structured identifiers, as reported by executable test suites. Layer L3 has no recall figure.
≈0.39 Mcharacters redacted per second≈10⁵ × human typingThroughput of the redaction engine, on the order of a hundred thousand times faster than a person typing.
339GDPR personal-data typesall 15 of Art. 9Including all 15 Article 9 special categories. Detector catalogue of the Pro and Team tiers.
03Reserved slotyet to be made
04Reserved slotyet to be made
05Reserved slotyet to be made
06Reserved slotyet to be made
How to join
Step onto the belt. No forms: just write to us.
Tell us what you want to build and which station you see yourself in. A person on the team will reply.