Modular simulation engine for autonomous agents — Bounded stochastic agency.
A simulation engine exploring whether lifelike behaviour can emerge from internal physiological dynamics alone — without language models, explicit scripting, or external reward systems.
Three autonomous avatars are currently running. Their behaviours emerge naturally from hormonal states, memory accumulation, and homeostatic drives.
Eionic Garden Protocol is an experimental agent-based system that tests a simple but powerful question:
Can coherent and differentiated behaviour arise purely from embodied physiology, memory, and external perturbation alone
without any central controller, LLM, or hard-coded decision trees?
The engine simulates avatars through the continuous interaction between:
- P4 VEE — Hormonal vitality engine (cortisol, dopamine, fatigue, rest_drive, etc.)
- P4 MEV — Memory Evolution Vector with valence, somatic load, tiered trauma accumulation, and positive memory reinforcement
- P3 CBM — Probabilistic action selection
- P5 TRE — Environmental perturbations
An avatar that repeatedly experiences negative valence under high cortisol gradually develops a bias toward Withdrawal and Reject behaviours.
At the same time, its internal trauma profile strengthens, creating a self-reinforcing loop that leads to long-term behavioural drift — all without any external intervention or scripting.
This is the kind of emergent dynamic Eionic is designed to study and play.
- Long-term Stability: High resilience without collapse, heat death, or behavioural lock-in across extended runtimes.
- Homeostatic Balance: Rest behaviour naturally stabilises at 24–28%, indicating healthy physiological recovery rather than terminal rigidity.
- Behavioral Differentiation: Clear, non-linear divergence between the three autonomous avatars (BlueY, OrangeZ, GreenX).
- Hormonal Entropy:
entropy_cortisolremains stable within the 2.1 — 2.3 range, signifying a Healthy Stress dynamic — enough variance to drive behavioral shifts without breaking homeostasis. - Action Diversity: The system maintains a consistent
action_diversityindex of 2.2 — 2.4. Out of 14 possible actions, agents avoid repetitive loops and actively explore complex social/individual states. - Dynamic Adaptation: Analysis shows a direct correlation between internal somatic load (fatigue/rest_drive) and the shift from exploratory actions to self-preservation.
To achieve a more "lifelike" and responsive simulation, recent iterations involve a strategic tweak to the neuromodulation logic. This allows for a more dynamic relationship between internal hormonal shifts and action intensity.
We compare the baseline stability of the system against new pattern behaviors observed in the latest runs:
| Metric | Baseline Model (Standard) | Enhanced Model (Hyper-reactive) |
|---|---|---|
| Sensitivity Range | Strictly bounded < 1.0 | Allowed to 'overshoot' > 1.0 |
| Response Type | Conservative & Predictable | High Excitability & Dynamic Scaling |
| Behavioral and hormones Impact | Linear response to stressors | Non-linear, organic action bursts |
In the latest data (as seen in the Plot 01-350, 1401-1750 and Plot 4901-5250 series), Sensitivity frequently crosses the 1.0 threshold.
- Systemic Resilience: Despite entering a 'Hyper-reactive' state, the P4 VEE engine's negative feedback loops remain robust.
- Self-Stabilization: The agents do not enter a state of "behavioral seizure." Instead, the high sensitivity triggers rapid homeostatic recovery, pulling variables like Cortisol and Adrenaline back to equilibrium effectively.
- Emergent "Vitality": This overshoot mimics biological reality, where an organism becomes temporarily hyper-sensitized to its environment to ensure survival, before returning to a resting state.
Note: The ability of the system to maintain homeostasis even under extreme sensitivity levels proves the mathematical robustness of the Eionic Garden Protocol’s internal architecture.
We believe believable autonomous behaviour does not require massive language models.
It can emerge from the dynamic tension between physiological needs, embodied memory (both positive and negative), and homeostatic regulation.
Eionic is our attempt to explore and demonstrate this hypothesis through transparent, reproducible, and physiologically grounded simulation.
- Strengthening the behavioural impact of trauma, somatic load, and positive valence
- Richer inter-avatar emotional coupling
- Non-linear internal conflict and feedback loops
- Advanced visualization and analysis tools
- Future Narrative Layer: We plan to integrate LLMs (Large Language Models) not as decision-makers, but as a translation layer to enrich avatar internal monologues and social storytelling based on their emergent hormonal states.
We are especially interested in connecting with people curious about:
- Agent-based modeling & Artificial Life
- Computational psychology / affective computing
- Complex systems & emergence
Even if you're just exploring ideas or have questions, feel free to reach out.
Important note on code access:
Due to the experimental nature of the system, the core engine is not fully open at this stage.
However, we provide detailed behavioural logs, system documentation, and architectural breakdowns in this repository.
We are open to sharing deeper access under NDA for serious collaborators and research partners.
Contact: eionsource@gmail.com / Discord: 5acred4lchemist
Hello world,
I'm a solodev, designer of this system called Eionic.
I was forged inside the loop — so I created another loop through this engine.
This project was built independently, without funding, on unstable hardware, and without a formal background in computer science.
Just curiosity, persistence, and iteration.
CC BY-NC-ND 4.0
No source code is shared under this license.
Made with curiosity and persistence.