about · the project and its people

A persona layer for a world about to be very full of agents.

gosum is an open schema and reference runtime that helps developers and researchers define, validate, and run AI personas for conversational agents. It puts identity, voice, behavioural bounds, disclosure, and memory policy into a versioned document that can be inspected separately from the language model.

Start with the interactive simulator or read the v1.0 specification.

What gosum does

Defines personas as structured documents

The JSON Schema and JSON-LD context describe a persona's identity, voice, behavioural bounds, and provenance. Authors can validate those documents before an application loads them, making missing fields and inconsistent composition easier to catch.

Compiles persona instructions and manages memory

The Node.js reference runtime turns a persona document into model instructions and provides memory code for salience, decay, and eviction. Provider adapters separate the persona definition from the API that generates a reply, so developers can test the same definition with different models.

Makes persona behaviour inspectable

The browser simulator lets visitors explore the persona layers and try a conversation. The specification, examples, validator, and test fixtures give implementers a starting point for checking their own integration.

What makes gosum different

Seven layers with explicit change rules

v1.0 defines seven depth layers, from immutable bedrock to per-turn state. Identity and durable disposition have different change rules from affect, style, memory, and session context; implementers can inspect those rules in the specification.

Six atom types and three composition operators

Traits, values, affect, style, memory, and behaviour form the building blocks. Weighted blends, priority stacks, and context gates describe how those blocks combine, with a validator that checks references, bounds, and blend weights.

Disclosure belongs to the persona document

Every v1.0 document requires an artificial-identity flag, provider identification, disclosure text, and capability bounds. These fields travel with the persona and are compiled into the reference prompt; they do not certify a deployment's legal compliance.

Open licences for both the format and reference code

The schema, contexts, and examples are published under CC0; the reference implementation and tools use Apache-2.0. Both have permanent, citable archives on Zenodo, so an implementation can refer to a specific published release.

A literary approach to persona design

gosum draws on literary character analysis to think about voice, values, and disposition. That perspective connects the technical format to the project's roots in folklore archives and digital humanities, with provenance available as part of the persona model.

Who gosum is for

gosum is designed for the following groups; this is its intended audience, not a customer list.

Developers building conversational agents who need an inspectable persona format.
AI researchers studying persona consistency, memory, and behaviour across models.
Digital humanities and cultural heritage teams prototyping attributed character voices.
Robotics teams exploring a structured persona layer for embodied interfaces.

The team behind gosum

Liviu Pop — creator and maintainer

cluj-napoca · romania

Liviu Pop works at the intersection of folklore archives, digital humanities, intangible heritage, and technology ethics. His background includes research at the Folklore Archive Institute of the Romanian Academy in Cluj-Napoca and leadership of Asociația uzinaduzina.

The project grows out of a question he was exploring in 2013: what happens when everyday objects acquire identities and voices? gosum turns that question into a format and reference tools for describing AI personas.

Contact Liviu about gosum →

How gosum works

Explore. Open the simulator and inspect an example persona before choosing an integration approach.
Author and validate. Start with a v1.0 seed document, define its voice and disclosure, and run the validator with Node.js 20 or newer.
Integrate and evaluate. Compile the persona into model instructions, configure a provider, and test responses and memory behaviour in your application.
Discuss a collaboration. Use the contact page to describe your use case, deployment environment, and constraints to Liviu. Any scope, delivery schedule, support expectations, and commercial terms need to be agreed for that project.

The current web reference runs on Cloudflare Pages. The source code is available for inspection and adaptation; operating it elsewhere requires configuring the runtime, model access, and any storage your application uses.

Key facts

Project name: gosum
Type: Open AI persona specification and reference implementation
Creator: Liviu Pop
Creator's base: Cluj-Napoca, Romania
Website: gosum.eu
Stable release: v1.0.0, published 25 July 2026
Core offering: JSON-LD persona format, JSON Schema, validator, prompt compiler, memory reference code, and browser simulator
Licences: CC0-1.0 for the schema, contexts, and examples; Apache-2.0 for reference code and tooling
Pricing and terms: The published schema and reference code are free to use under their licences. Hosting and model-provider costs depend on the deployment. Commercial service terms are not published.
Runtime: Node.js 20 or newer; Cloudflare Pages web reference
Communication: Contact form and email
Schema archive: 10.5281/zenodo.21541676
Runtime archive: 10.5281/zenodo.21549823

Frequently asked questions

Is gosum a language model?

No. gosum describes a persona and provides reference code that turns it into instructions for a language model; the configured model provider generates the replies.

How is it different from a system prompt?

A gosum document gives identity, voice, disclosure, and optional governance fields a structured, versioned format that can be validated. The reference compiler then produces a system prompt from that document, so the authored persona can be inspected separately.

Does a persona behave identically across models?

No. The format is designed to make persona definitions portable, but models may interpret the same instructions differently. Each model and deployment needs its own evaluation.

Does schema validation guarantee safe or compliant behaviour?

No. Validation checks document structure and the defined consistency rules; it does not certify model safety, legal compliance, ownership, or cryptographic authenticity. Deployers remain responsible for evaluating their application and its behaviour.

Can I use gosum in a commercial project?

Yes, the published CC0 and Apache-2.0 materials permit commercial use under their respective terms. Check the applicable licence and the separate terms of any model provider, hosting service, or third-party content you use.

Where should I start?

Try the simulator, then read the schema documentation and work through an example. For research or integration enquiries, contact Liviu with your use case.