A persona layer for a world about to be very full of agents.
What gosum is
gosum is two things held together. It is an open standard for personas — the layered, composable artefacts that give an AI agent a voice, a memory, and a sense of where it is from. And it is a governable runtime that loads one of those artefacts and behaves accordingly.
A persona, here, is built from six kinds of atom — trait, value, affect, style, memory, behaviour — composed by three operators and arranged across seven layers of depth, from a fixed core that never changes to a surface that adapts each turn. The deeper the layer, the slower it is allowed to move. That gradient is the whole safety story: a persona can live and gather experience for years without ceasing to be the persona that was signed off.
A persona is not a system prompt and not a configuration. It is a small, signed library of voice, lineage, and accumulated memory, written in a vocabulary anyone can read and any runtime can load — portable across models and hardware, versioned by a content hash, auditable from the first turn to the last.
Why it exists
In the coming decade most software you talk to will need a voice, and most robots will need a small story about themselves. Today that voice is improvised — a long system prompt, tweaked weekly, owned by no one in particular, drifting under model updates and impossible to audit.
gosum proposes the opposite: that personas are first-class artefacts — sourced from human literary tradition, governed like any other sensitive system, and portable across runtimes. Less improvisation. More craft.
Sourced from the page
Most AI personalities are a handful of trait scores or a paragraph of instructions. gosum's are distilled from the human literary tradition instead — the centuries of fully realised characters that no synthetic dataset can match — read by the same humanities methods long used to study them, and rendered into portable, modern voices. A character is not copied; its transferable structure is distilled, attributed, and set to work in a new role.
Who’s behind it
Researcher at the Folklore Archive Institute of the Romanian Academy in Cluj-Napoca and president of the Asociația uzinaduzina. He works at the intersection of folklore archives, digital humanities, intangible heritage, and the ethics of contemporary technology — and wrote the prequel to this project, on objects gaining identity, back in 2013.
Where it is, today
The schema is open and free to read and use, published as v0.1 with a v0.2 memory layer. The simulator is live: a small cast of personas who remember what you tell them, keep the vivid memories and let the faint ones fade as time passes, and forget you entirely on request. Two working papers place the idea in both artificial intelligence and the humanities. The standard is the open part; the runtime, the governance, and the audit are the commercial layer — the part that needs operations. See the architecture →
“In the coming years, the robots around us will not just perform tasks; they will tell stories, share jokes, offer comfort — each in their own style.” liviu pop · 2013 · ipv6 forum