Open standard · Character infrastructure
Protagons
Portable character identity for any AI model.
- Role
- Creator of the standard
- Type
- Open standard + API
- Timeline
- 2025
- Live at
- protagons.com
The friction
AI personas live as prose system prompts. You can’t diff them, fork them, or move them between models without hand-rewriting — and the reasoning walks out the door when the prompt engineer does. Two products that need one voice means copy-paste-and-pray. Prose has no source of truth.
Where AI helps
A pipeline of 28 extraction agents reads a description or real writing samples and produces a structured, evidence-backed, confidence-scored character file — around 287 fields in about two minutes — that a human would never hand-author. The payoff is portability: compile the same identity to any model and the character holds.
Who it helps — Developers and builders wiring characters into products — agents, NPCs, brand-voice systems, TTS pipelines. Not end-user writers; that’s USAW, built on this standard.
What I built
-
An open schema
JSON Schema 2020-12: 28 dimensions — linguistic and psychological — across 287 fields, with provenance and safety tiers.
-
A compiler
One
.protagon.jsonbecomes a system prompt, SOUL.md, ElevenLabs voice, or Custom GPT; recompile per model. -
A public library + free BYOK API
Generate, chat, validate, and deploy characters others can build on.
Proof