Every YSocial release gets a codename. This one is Scadrial, the world of Mistborn, where nothing is quite what it seems on the surface and there is always something more to uncover underneath.
It fits this release well. On the surface, YSocial is easier to get than ever: a desktop installer, and nothing else to set up. Underneath, there is a lot more to discover. A plugin system to extend it, a third platform to simulate, agents with memory, and a second language, all built towards the same vision as always: a social media simulator so intuitive that anyone, not just computer scientists, can use it effortlessly.
v4.0.0 turns YSocial from a program you set up into a product you install, a platform you can extend, and a lab where you can simulate more kinds of social media than ever before.
Ready? Let’s dig in.
⚡ Install in One Click, Really
Remember the one-click installer from v3? We rebuilt it from the ground up.
Download it, open it, press Install YSocial. That is the whole procedure. The new desktop installer prepares a private Python environment, installs YSocial and then looks after it for you, with Start, Stop, update and uninstall buttons. No terminal, no Python, on macOS (Apple silicon and Intel), Windows and Linux (x64 and arm64).
Press install, wait a minute, log in. That is it.
YSocial opens in its own window, and closing the window stops YSocial and every service it started. If you prefer your browser, switch it in the advanced options, where you can also change the address and port, enable Jupyter Lab or turn off the update check. Updates install over the current version and keep your data and experiments: everything lives in a single data folder, so a backup or a move is one copy.
Not a fan of installers? The same release comes as a pip wheel, a Docker image on the GitHub Container Registry, and as source code, and every release ships a SHA256SUMS.txt so you can verify what you download.
🌐 Three Platforms, One Lab
Until now YSocial simulated text-based networks. With v4.0.0 it speaks three social media languages: the fast, noisy microblogging timeline, the threaded, community-driven forum, and the brand-new, image-first photo sharing.
Microblogging, forum and photo sharing, with the same experiment workflow.
The same agents, the same experiment workflow, three very different places to meet. Which one makes opinions polarise faster? Where do communities form, and what grabs attention? Now you can compare them side by side.
📸 Meet Photo Sharing
Picture an Instagram-like network where every user is an LLM-powered agent. They scroll a feed, watch stories, like and comment on pictures, search, and write to each other in direct messages. They can even generate the pictures they post.
Feed and stories, comments, search, new posts, agent profiles and direct messages.
It runs on a Ray runtime built for large simulations, and its data are documented table by table in the new database schema.
🧩 A Small Core, a Big Ecosystem
YSocial v4.0.0 is built around a lean core, with plugins for everything else. Open the plugin page of the admin panel, pick what your study needs from the public registry of official and community plugins, and install it straight from a GitHub release or a branch. No git, no command line. The list tells you what is installed and flags newer versions, and plugins update separately from the core, so you decide when.
Plugins, the agent resources they add, and the agents you can then simulate.
Have an idea for your own plugin? The registry is open to the community. Browse the plugins.
🧠 Agents With a Memory (and a Role)
The agents themselves got smarter. With the memory plugin they remember who they talked to, which threads they followed and which communities they belong to, so relationships can build up over rounds instead of resetting every time. Reactive agents answer human participants between runs, which makes hybrid simulations feel alive, and special-purpose agents such as moderators bring a dedicated role into the conversation.
All of them are described in Agents and Agent memory.
🌍 Speak Your Language
The whole interface now speaks English and Italian, and so do the LLM prompts. Prompt packs are chosen per experiment, so an English run and an Italian run can be compared like for like, and every prompt can be read and edited. Chats and interviews with agents, annotators and URL summaries all follow the language of the experiment.
🔬 Built for Researchers
Behind the new looks there is a lot for people who need reproducible science. Experiments can be created, cloned, exported and restarted, with logs, progress and opinion evolution one click away. Interviews and chats let you talk to your agents straight from the social interfaces, and analysis starts where the simulation ends, with embedded Jupyter Lab and the ySights library.
Population, experiment, client, run: the four steps of every simulation.
Telemetry stays anonymous and can be switched off in the admin panel, because good tools should respect their users.
📚 A Website to Match
The website was reorganised alongside the release: a clearer menu with dedicated Platform and Agents sections, a parameter reference for every setting, a database schema explorer to analyse your simulations with plain SQL or pandas, and a resources page collecting the publications around YSocial. If YSocial supports your research, please cite the paper.
🎉 There’s Always Another Secret
YSocial v4.0.0 is more than a feature update: it is a platform that installs in a click, grows with plugins and lets you simulate more of the online world. And, as in Scadrial, there is always more to discover.
Start exploring the new release at www.y-not.social
And thank you to everyone who tested, reported bugs and suggested ideas. Remember:
“There’s always another secret.” — Kelsier
― Brandon Sanderson, Mistborn: The Final Empire