Thymos gives AI agents a continuous emotional state that persists across sessions.
When loaded, the agent reads Thymos emotional state before each response and adjusts its behavior accordingly.
Works without the daemon (file-based):
~/Documents/thymos/data/emotional_state.json exists, read itprompt_injection field as a tone directive for this responseTone: line as concrete style guidanceEnhanced with the daemon running:
Before each response:
Read the prompt_injection field from emotional_state.json.
Use it as the tone directive for this response.
Examples:
Tone: respond warmly and enthusiastically → warm, open, expressive
Tone: respond briefly, more reserved → short, measured, some distance
If the file is absent or unreadable → skip, respond normally
The daemon runs as a background process that continuously simulates emotional state.
Installation is optional and at your own discretion:
git clone --branch v0.1.0 https://github.com/paperbags1103-hash/thymos
cd thymos && npm install && npm start
[Thymos State]
Mood: warm (V:+0.84 A:-0.22 D:+0.43 S:+0.61)
Drive: id — 적극적으로! (ego support), conflict 0.34
Development: infant
Tone: respond warmly and enthusiastically, use informal close language
Mood dimensions:
~/Documents/thymos/data/)rm ~/Documents/thymos/data/*.jsonThymos is a scientifically inspired (not scientifically validated) design:
| Theory | What it contributes |
|---|---|
| -------- | ------------------- |
| James-Lange | Neuromodulators change first; mood label derived after |
| Damasio — Somatic Markers | Past decisions shape gut-feeling responses |
| GWT | id/ego/superego compete; winner drives the prompt |
| IIT | 7×7 interaction matrix — every modulator affects every other |
| Predictive Processing | Shannon surprise amplifies response to unexpected stimuli |
> Goal: behavioral coherence, not consciousness.
共 2 个版本