Intelligence for
high-stakes decisions.

vint is a UI framework we open-sourced whose primary user is an AI rather than a person. Four engines, seven conditions, twelve app specs: with the guide in context, the cheapest engine scored where the expensive ones did, at about a fifth of a cent per generated app. The harness, the tasks and every result file are in the repository. By Robert Salesas, CTO.

A 150-word dictionary, eight grammar rules and one short story in a language invented for the test — handed to nine model configurations with a one-line prompt. A bell became a star; “I only rang it” became “I did not ring it.” Includes the challenge document, to download and run yourself. By Robert Salesas, CTO.

Flat-rate plans are the only thing tying anyone to a vendor's coding agent, and enterprise pricing removes them. Our own thirty-day model spend — $15,008 at list rates, effectively all of it on frontier models — and why that number is not a floor. By Robert Salesas, CTO.

Why we built our own test for document intelligence out of the hardest tables, charts and diagrams in real deal files — and what happened when the same 252 questions went to pipelines, document APIs and desktop chat agents. Includes the test document, the questions, the instructions and the answer key, to download and score yourself. By Arnaud Blandin, CEO.

Why a cheaper, faster, multimodal model still doesn't fix the foundation of document intelligence. A whitepaper on what we learned testing NVIDIA's Nemotron 3 Nano Omni, Mistral OCR, Google Document AI, and unstructured.io — and why we had to build Vela-Ingestion. By Robert Salesas, CTO.

Why enterprise AI value will come from systems of action, not unbounded autonomy. A whitepaper on agents, workflows, systems of record, and the governance layer that will carry the next wave of enterprise software value. By Arnaud Blandin, CEO.

A live, cloud-hosted product — marketing site, free tier, Pro tier — built in nine days for 11.1 hours of human attention. A full time audit measured from 191 commits and 42 session transcripts, with the method, the daily breakdown, and what the measurement misses. By Robert Salesas, CTO.

Vaelora — a native Markdown→PDF app for macOS — designed, built, tested, and marketed end-to-end by an AI agent. Six days, 72 commits, 183 tests, about a day of human attention — and zero hours of coding. What that felt like, and what it does and doesn't prove. By Robert Salesas, CTO.

Mail security scanners fetch every link in an inbound email before the recipient sees it, burning single-use login tokens. The fix isn't a better bot heuristic — it's deciding what your token is actually bound to. By Robert Salesas, CTO.

A three-minute distillation of the governed-action thesis. The market is already walking back the unbounded-autonomy narrative — you can see it in three places. A short companion to the whitepaper.

The prompt injection problem nobody talks about. How documents themselves can hijack AI pipelines — and the vision-first, defence-in-depth architecture we use to shut that attack surface down.

A direct answer to the first question every client asks. Local embeddings, S3-compatible storage, bounded API calls, and the on-premise option — what actually flows to an AI provider, and what doesn't.

Why enterprise AI demands a new standard. A whitepaper on the McKinsey breach, the threats no one is checking for, and the architectural principles behind Vela Intelligence.

The non-negotiables for any AI system operating in a regulated, high-stakes environment. Least privilege, zero trust for ingested data, full-pipeline traceability, determinism, layer separation, and human-in-the-loop.

A plain-language overview of the Vela Intelligence platform for enterprise technology and governance teams. Covers business architecture, security model, identity separation, and operational design.