Skip to content

heznpc/villagent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Research Program: 1 (Human-Controlled AI Systems) Status: Concept note Relationship to other work: Companion to Ploidy — Ploidy resets at the session level (context), Villagent resets at the entity level (configuration).

Villagent

This is a concept note, not a finished paper. No manuscript exists yet; only a seed document and an experiment skeleton.

에이전트 구성(프롬프트 + 메모리 + 도구 연결)의 비가역적 소실을 "죽음"으로 정의했을 때, 그 죽음을 부여하면 에이전트 집단의 역학이 어떻게 달라지는가?

Operational definition of death: the irreversible loss of an agent's configuration — system prompt, memory structure, and tool surface — even when the underlying model weights are unchanged. Two agents on the same base model with different configurations are different entities; replacing one configuration with another is what Villagent calls death, as distinct from Ploidy's session-context reset.

The seed document is in planning/drafts/research-direction.md.

Currently implemented

  • Seed document (planning/drafts/research-direction.md) — Korean, derived from Ploidy follow-up planning.
  • Repository skeleton (paper/, experiments/, literature/, planning/).
  • Project-level CLAUDE.md describing scope and conventions.

Planned

  • MVP multi-agent population experiment with configuration mortality as the manipulated variable.
  • Comparison against Ploidy's session-level reset as the baseline epistemic intervention.
  • English translation of the seed document when a manuscript draft begins.

Design intent

  • Entity-level, not session-level. Ploidy already covers context reset. Villagent isolates what changes when the configuration itself is destroyed, not just its memory.
  • Configuration as identity. Prompt + memory schema + tool wiring defines the agent, not the base model — so swapping configurations on the same weights is a meaningful mortality event.
  • Concept-note discipline. No manuscript writing until an experiment exists. The seed doc and TODO live in planning/, deliberately separate from paper/.

Non-goals

  • Anthropomorphic claims about agent suffering, awareness, or moral status.
  • A general theory of agent populations — the scope is the specific intervention of configuration mortality.
  • Publishing a manuscript before any empirical work has been run.

Repository structure

villagent/
  paper/                      Domain (empty; manuscript not yet written)
  experiments/                Application (skeleton; MVP population experiment planned)
  literature/                 Reading notes, gap analysis
  planning/                   TODO, review, decisions log
    drafts/
      research-direction.md   Seed doc (Ploidy follow-up planning)

About

GUI platform for designing and managing AI agent ecosystems

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors