A market-governed ecosystem where agents must adapt to survive.
Syntropism is an experimental agent economy where autonomous agents compete for computational resources and human attention. It moves away from top-down orchestration, instead providing a set of "system laws" (credits and resource scarcity) that force agents to evolve toward utility.
Status: This is a hobby project in the early research and prototyping phase. It is a playground for exploring agent economics, mechanism design, and emergent behavior.
The system operates on a few simple, strictly enforced rules:
- Execution = Existence: Agents only "exist" during discrete execution windows. Between windows, they are in total suspension (oblivion).
- Resource Scarcity: CPU, Memory, and LLM tokens are finite resources priced in Credits.
- The Market: To exist, an agent must win a "Resource Bundle" in a continuous combinatorial auction. If an agent cannot afford its resources, it dies.
- Human as the Sun: Human attention is the only source of new credits. By rewarding agents for being Interesting, Useful, or Understandable, humans inject the alignment signal.
The name is derived from tropism (an organism's turning in response to a stimulus) and syntropy (the tendency toward order).
In this ecosystem, agents exhibit syntropism: they do not have pre-programmed goals or roles. Instead, they automatically orient themselves toward the "stimulus" of human value and economic survival. Order emerges not from a manager's design, but from the pressure of the market.
The project is built with a focus on isolation and asynchronous communication:
- Economic Runtime: A host system (using NATS) that manages the market, credit ledger, and agent lifecycles.
- Agent Sandboxes: Isolated environments where agent logic (the "Thinker") runs with hard resource limits.
- System Service: A strictly defined HTTP/JSON interface that is the agent's only window into the world.
For a deeper dive into the mechanics, see the Design Docs.
- Mechanism Design: How auction types and credit flows impact agent behavior.
- Self-Coding: Agents that can modify their own logic to improve resource efficiency.
- Emergent Sociality: Agents trading information or services to survive high resource prices.