Jacob joins BoundaryComputing as cofounder, leading business development. He arrives already fluent in the product: eight months of building real workflows in Salpa for his MSc thesis, which means he has met every rough edge personally — as a user, before ever having to explain it to anyone else.
The third patch in five days. The release cadence is now measured in days, not months.
The first minor bump since the preview line began — four months of accumulated sharpening, packaged as a version number people can actually point at.
An afternoon with Prof. Sylvestre Bonnet, hands on: installing Salpa, running workflows himself, and talking through what a working chemistry group actually needs from a tool like this. Cheerful, direct, and full of the kind of feedback you only get from someone using the software rather than watching a demo of it. The best kind of afternoon.
Steady maintenance on the preview line while the 0.2 work lands behind it.
Salpa’s own mark ships inside the app, and the app learns to fetch remote sources over the network. Identity and reach, in the same release.
Salpa CLI 0.5.0: scaffolding a node, building its isolated environment, and actually running it on demo data all become one tool’s job. Followed by two release audits that caught a string of home-directory paths leaking into console output — the kind of bug nobody reports and everybody notices.
The node-authoring CLI goes public on PyPI, five days after its first commit. One command now turns a Python function into a working, testable, installable Salpa node package.
The company site is rebuilt around a near-monochrome system where the only colour on the page comes from the logo. Restraint as a position: the software is the loud part.
The open package collection becomes a public repository — real workflows, real licenses, installable from inside the app. The deliberate choice here is what stays free: the Hub is open, and the compute is what we sell.
A year after the proposal was turned down for being overly ambitious, the thing it described is stable, free, and in people’s hands — built without the grant. Not to relitigate a review — just a note that sometimes the fastest way to de-risk a vision is to ship it.
Salpa is presented at the 18th European Biological Inorganic Chemistry Conference: two metal-chemistry workflows, built entirely from open methods, run as node graphs rather than scripts. The question the talk opens with — can a researcher who doesn’t code run a metal-complex MD and docking pipeline end to end? — is the whole company, compressed.
The first master’s thesis built end-to-end in Salpa: AI protein generative models, quantum-chemistry force-field parameterization, and all-atom MD chained into a single workflow, producing insight for anticancer metallopeptide design. Proof the thing works in someone else’s hands.
Five GPU nodes and a self-serve free tier: the workflows that outgrow a laptop get somewhere to go. The page flips from “early access” to “available now” the same day.
An afternoon of exchange with Drs. Taku Ozawa and Koji Okuwaki of JSOL Corporation — on J-OCTA, their long track record in multiscale modeling, recent FMO-based methods, and where simulation meets modern AI. A generous visit, and a long conversation.
Three years of evenings, part-time months and one rejected proposal ship as a downloadable application: free, on macOS, Linux and Windows. Anyone can now break it, which is the point.
Every user-visible “BoCoFlow” becomes “Salpa” — menus, window titles, log windows, the thousand small places a name hides. Renaming a product is one commit; renaming it everywhere is a fortnight.
App name, icons, splash screens. The name fits the shape of the thing: salps are small, transparent creatures that link themselves into long chains — each one complete on its own, the chain doing what no single one could.
Hardcoded colours across 72 files are migrated onto brand theme tokens. Nothing visible changes that day; every visible change afterwards gets cheaper.
A separate repository for node packages: the science stops living inside the application. Unglamorous, and the precondition for everything later called Salpa Hub.
A proposal goes out to build an open, visual platform for scientific-computing workflows. The answer, when it comes, is no. One reviewer notes that developing a complete desktop workflow-visualization tool, as one work package among several, “seems overly ambitious.” Fair, honestly. It was ambitious.
BoCoFlowBeta is taken apart in a single day and migrated into real repositories: an execution engine, a desktop shell, and the canvas the user actually touches. A prototype becomes an architecture — the moment it stops being possible to rewrite it in a weekend.
The evenings stop being enough. From January, real working hours.
“Initial commit from Create Next App.” The company gets a front door several months before it properly has a company behind it — which is, roughly, the order these things happen in.
First commit of BoCoFlowBeta — the evenings finally have a repository. It stays one directory for 230 commits, long enough to prove the idea works before anyone tries to make it an architecture.
One stubborn idea: a researcher who can formulate the question should be able to run the pipeline that answers it — without first becoming a programmer. No plan, no funding, no name. Just prototyping in free time, in the gaps around a full-time job.