EBONEXEBONEX
Sign up
Changelog

What changed, and when

Every release that reached researchers, newest first, written in the words you would use rather than the words the commit used. If something you rely on changed, it is on this page.

Open EBONEX See the guides

A website that shows the real product

The site shows the app, not a drawing of it

The home page used to carry an illustration of the product. It now carries real screenshots of the running app, and the guides are written from real studies: the docking guide is a transcript of an actual run, with the screens and the numbers that run produced. There is a new About page, and signing in from the site now lands you inside the app instead of returning you to the marketing page.

Developer documentation and this changelog

A Docs page for anyone who wants to build for EBONEX: what Tools, Skills, Plugins and MCP servers are, the exact manifest formats with their real limits, how to connect an MCP server, and how to publish to the Marketplace and get through review. And this page, so you can see what changed without asking.

Figures with spaces or brackets in their names show up again

A figure whose filename contained a space or a bracket silently showed nothing in chat and in the report viewer. Those images now render, Results lists every figure that is actually on disk, and a report opened from a folder path finds the figures beside it.

Molecular dynamics: failed runs are listed and never charged

A molecular dynamics run that does not finish now stays in your Jobs list with the reason written on it, instead of disappearing, and you are never charged for it. The rented GPU machine is shut down on every path out — finished, failed, timed out or cancelled — so nothing keeps running after a job ends. Protein-ligand simulations are still being worked on; this entry will say so when one completes end to end.

A slow start explains itself, and a dropped connection comes back

A workspace that takes a long time to wake now says so and keeps retrying by itself instead of hanging on a blank screen, and a live connection that drops reconnects instead of leaving the history stuck.

You can watch a run's steps while it works

A running study shows its steps as they happen, and retries that finished in a second no longer clutter the transcript. On a narrow screen the chat and Results columns stay inside the page, and a project's chat count no longer counts empty drafts.

Reference checking holds up better

The reference checker now resolves a few references at a time and retries a lookup that got dropped, instead of failing the whole batch when one service is slow.

Everything a simulation can tell you, and papers that read like papers

Full analysis after every molecular dynamics run

A simulation used to end at RMSD, RMSF, radius of gyration and clusters. Every run now also measures per-residue SASA, hydrogen bonds over time for the protein and the ligand separately, DSSP secondary structure, principal component analysis with its scree plot, and the PC1/PC2 free-energy landscape — and, when there is a ligand, its RMSD, its distance from the pocket, an interaction fingerprint and, where the system allows it, MM-GBSA. Each one writes a figure and its tables, with a caption and a plain reading of what it shows, straight into the report.

A manuscript that looks like a published article

The paper the writing tool produces now takes the shape of a journal article rather than the output of a program. There is no bulleted limitations section: the limits close the Discussion as prose, each one naming the conclusion it blocks. Acknowledgements are included, the software that actually ran is cited with its version, and a docking report states its re-docking control from the evidence rather than from a template.

Results, Library and Images open instantly

These pages now draw what they showed you last and refresh behind it, so there is no wait before you can read. Library shows each folder as soon as it is read, Images loads sixty at a time, a whole row is clickable rather than just its name, every file type has its own icon, and a project you have finished with can be archived.

Tell me when it is done

Every running job has a Notify me beside its stage, so you can close the tab and be told when the work finishes.

Super Brain is easier to steer

Reset view and auto-rotate are proper buttons on the canvas, the knowledge map has a legend of note kinds, and note previews read as plain text instead of showing raw formatting marks.

One design for Plugins, Tools, Skills and MCP

All four lists now use the same row: status, an on/off switch and Delete, with any control the item cannot take disabled and the reason on hover. A Tool no longer appears on two pages, tools say which plugin they came from, and skills are listed by their readable name rather than their identifier.

Marketplace, Agents and Schedule

Every Marketplace item now has its own page with Overview, Access and Publisher, and you publish under your own account with your own detail tabs. The Agents page marks the built-in teams and lets you edit or delete your own, and adds a Review article team. The Schedule page gained templates and a month grid that fills the page.

My Computer

The desktop has the EBONEX picture as its wallpaper, drawn for the exact size of your screen, and a file you reopen shows what is on disk now rather than what it held when you last looked.

Launch — EBONEX is live on ebonex.co.uk

The product is open on its own domain

EBONEX runs at ebonex.co.uk. Sessions already signed in on the old address keep working.

Molecular dynamics on a GPU rented for your job

The machine EBONEX runs on has no GPU, so simulation could not run at all. Now every simulation rents a GPU machine for that one job: it picks the cheapest card that passes its own checks on memory, reliability, driver and bandwidth, sends the work, streams the progress back to you, brings the results home, and destroys the machine afterwards — whether the job finished, failed, timed out, hit its spend cap or was cancelled.

Super Brain remembers you, and saves what worked as a skill

One personal memory for your whole account, shared by every project: who you are, what you have taught it, and a card for each project. The agent writes to it as it works and reads it at the start of every conversation. When a procedure works, it offers to save it as a skill in your own library, and anything it saved on its own is marked unreviewed until you say otherwise.

The agent can work the app for you

Ask it to open a page, start a chat or change a setting and it does. It is not allowed to delete your account or change any limit, plan, credit or quota — those stay yours.

My Computer looks like EBONEX

Your Linux desktop used to look like a stock 2010 desktop. It now has a slim top bar with the date and time, a dock with Files, Browser, Terminal, PyMOL, a text editor, Results and Downloads, and dark windows throughout. The agent puts each study on the desktop where you can find it, and the desktop no longer steals the keyboard from the chat box.

Research teams

Create a team, invite people by e-mail or by link, and accept, decline, remove or leave from inside the app, with the leader holding the permissions. Profiles gained a cover image and a photo, and a Usage page shows your recent requests and GPU hours by day.

Results, Library and Images rebuilt

Every result sits in its own folder with a way back to the chat that produced it, collection pages remember how you like to view them, a Library folder opens into its contents with a path back up, result files open as tabs, and an archived chat you reach from its results offers to restore itself.

Poses you can look at

A docking pose opens inside its receptor with the pocket highlighted and the receptor drawn pale so the ligand stands out, and the 3D viewer now opens on machines without hardware graphics instead of refusing.

Connections

WhatsApp connects from a QR code shown live in the chat, with a fresh code if it expires, and Telegram connects in one click.

The app starts about three times lighter

The molecular viewer and the PDF reader were 18.4 MB of a 24.4 MB start-up and now load the first time you open them instead. What is left is compressed once per release, 11.1 MB down to about 6.9 MB.

A new look, and the first tutorials

A fuchsia-violet-navy palette across charts, usage bars, cover images and card glows, with text and navigation left clear of it. The public site gained its first tutorial page for docking, simulation and paper writing, readable without signing in.

Before launch — the product itself

A research agent that finishes the work

Chat, Your Workspace, Projects, Library, Images, Results, Super Brain, Schedule, Connections and the Marketplace, built so that you describe a question and get the finished study back.

Docking as a whole study

The agent chooses the mode, finds the binding site from the evidence rather than being told, runs blind, site-directed, induced-fit or ensemble docking, analyses the interactions, draws the 2D, 3D and contact figures, and writes the report with every reference checked and an AI-use statement.

Papers with a contract

The writing tool holds the manuscript to a structure, keeps one folder per study, renders the report to PDF and Word, and includes 3D figures and the binding-site residue table.

A Linux computer of your own

Every account gets a persistent Linux desktop with a Python science stack already installed — numpy, scipy, pandas, matplotlib, seaborn, scikit-learn, biopython and RDKit — plus Open Babel and the PubMed, Paper Search, Crossref and OpenAlex literature servers.

One approval for a study, and a refund when it is our fault

The agent looks before it quotes, prices the whole study once, asks once, and re-runs at no charge anything that failed on a setting it chose for you.

Sign-in, waking up, and long answers

Google sign-in with Gmail and Calendar on the web, a single e-mail form with errors in plain words, and a return to the page you came from. A sleeping workspace wakes on a branded screen rather than a black one, schedules fire on time without a chat open, and an answer that reaches the length limit carries on instead of stopping mid-sentence.

Something missing?

If a change broke the way you work, tell us

We would rather hear it from you than find it in a month. Write with what you were doing and what changed, and it reaches the people who made it.

Email the team Developer docs