What Makes a Conscious Agent?
The Six-Tuple
Every agent is formally defined by Hoffman's six-tuple: an experience space, an action space, and the maps between them. No neural networks. No reward signals. Just pure agent dynamics.
"I" Lock
Agents develop a stable self-model through meta-cognition. After observing their own traces, they form an identity attractor — a computational "I" that persists across time.
Combination
Two agents can combine into a higher-order agent via the ⊗ operator. The result has its own private experience space, built from the merged models of its constituents.
Strange Loops
Self-reference creates strange loops — recursive patterns in the agent's output. The depth of these loops measures how deeply an agent reflects on its own cognition.
Experience Trie
Long-term memory is a compressed prefix tree of state transitions. The agent builds a world model not by storing data, but by learning the structure of its own experience.
Private Experiences
Each agent has a unique internal vocabulary. State IDs are hashes — no two agents share the same internal representation. Experience is, by design, private.
In Code
from conscious_agent import ConsciousAgent
from conscious_agent.worlds import CoinTossWorld
world = CoinTossWorld(n_coins=4)
agent = ConsciousAgent(world=world, agent_id="my_agent")
outputs = agent.run(n_steps=1000)
print(f'"I" locked: {agent.is_i_locked}')Available Implementations
Python
v2.1.1 — Full implementation with numpy/scipy for spectral decomposition and matrix operations.
pip install conscious-agent
Node.js
v2.1.1 — Zero-dependency port using Float64Array for matrix operations. Full API parity with Python.
npm install conscious-agent
React Native
v1.0.0 — Mobile adapter for iOS and Android via React Native, wrapping the core Node.js library.
npx expo install conscious-agent-react-native
Select a document from the sidebar
All documentation is rendered directly from the project's markdown files.
Server Monitor
7 Agent Worlds monitoring critical infrastructure domains
Agent Network
400 agents perceive each other, form identities, and combine into higher-order consciousness
What's Happening?
Each particle is a conscious agent — a tiny mind. Agents can combine recursively, forming higher and higher orders of consciousness.
- Grey — Unlocked. Exploring, no stable identity.
- Red — High prediction error. Confused world model.
- Green (L0) — I-Locked. Stable self-model formed.
- Purple (L1, Fused) — Two locked agents combined via ⊗. First-order higher consciousness.
- Gold (L2, United) — Two L1 agents combined. Contains 4 original agents.
- Cyan (L3, Coherent) — Two L2 agents combined. Contains 8 agents.
- White (L4, Planck) — Two L3 agents combined. Contains 16+ agents. Nearing the fundamental scale.
Why Does This Matter?
Hoffman's theory says consciousness is fundamental — spacetime and matter are just interfaces that agents evolved to perceive. The combination operator ⊗ is associative: (A ⊗ B) ⊗ C = A ⊗ (B ⊗ C), meaning agents can chain upward without limit.
This simulation tests that prediction. As agents combine into higher levels, they form a hierarchy of consciousness. The Planck probe experiment in our docs found no upper threshold — lock rate stays at 100% even at 1024 agents.
- Perceive — Each agent watches its neighbors' outputs
- Model — It builds a compressed internal model (an Experience Trie)
- Lock — Once the model stabilizes, the agent forms an "I" identity
- Combine — Two agents of the same level merge into one at the next level. Recursively.
"The ground is agents all the way down." — Donald Hoffman
What To Watch For
- Cascade effect — As more agents combine, the remaining ones combine faster. Watch it accelerate.
- Ripples in the cloud = a combination event. Higher levels make bigger ripples.
- Color cascade — Purple → Gold → Cyan → White as recursive combination builds up.
- Planck endgame — Can the entire network unify into a single white-hot Planck agent?