The deterministic runtime that advances agents, worlds, state transitions, and events.
HEAVEN
Persistent digital agents with observable behavior across time.
HEAVEN builds persistent digital agents inside deterministic simulations, making actions, memory, and decisions inspectable instead of opaque.
HEAVEN
Persistent cognitive simulation for digital agents whose behavior can be inspected across time.
PlatformEDEN
The first controlled world built to make the HEAVEN architecture visible.
Validation worldArchitecture
Controlled ticks, state transitions, and reproducible runtime execution.
RuntimeMemory
Event-grounded continuity with provenance and controlled access views.
ContinuityTrace
Decision paths, event history, and replay artifacts for reconstructing behavior.
InspectionBoundaries
Scope, safety, and identity limits for controlled simulation.
ScopeUse Cases
Research, simulation, and audit contexts where continuity matters.
ApplicationsResearch
Memory, decision, movement, logic constraints, and bounded deliberation.
FoundationsAccess
Research, collaboration, and early technical discussion channels.
ContactPlatform
HEAVEN
Most digital agents are temporary, stateless, and difficult to inspect after they act.
HEAVEN explores the opposite direction: persistent cognitive simulation for digital agents that carry internal state, form memory from events, and expose the traces behind their behavior.
It is built around agents that do not simply answer prompts or perform isolated actions, but exist inside structured worlds and remain reconstructable across time.
The project is organized as an orbit of connected layers: a deterministic engine, structured worlds, persistent agents, event-grounded memory, decision traces, replay tools, safety boundaries, and research foundations.
Persistent agents with internal state, routines, memory influence, and traceable decisions.
Structured environments where behavior happens inside controlled spatial and semantic constraints.
Event-grounded memory that carries continuity forward without becoming opaque generation.
Inspection layers that make behavior reconstructable instead of merely believable.
Explicit limits and constraints keep simulation controlled, scoped, and auditable.
System distinction
HEAVEN is the platform vision. heaven-engine is the runtime layer. EDEN is the first controlled world built to make the architecture visible.
The platform vision for persistent cognitive simulation and inspectable digital agent behavior.
The runtime layer for controlled ticks, state transitions, and reproducible execution.
The first controlled world built to make the architecture visible.
Validation World
EDEN
EDEN is the first controlled world built to make HEAVEN visible: an environment where persistent agents can move, interact, generate events, form memory, and be inspected over time.
-
01
Controlled world
A scoped world surface gives agents a defined context for movement, state, and review.
-
02
Agent activity
Actions happen inside visible spatial and semantic constraints, keeping conduct tied to a concrete scenario.
-
03
Event history
Important changes become reviewable records rather than disappearing into a transient session.
-
04
Continuity layer
Relevant context can carry forward with source linkage and controlled scope.
-
05
Replay inspection
Review reconstructs the sequence so behavior can be inspected over time.
- Named environmentRooms, zones, and spatial references give each agent action a concrete place inside the world.
- Validation surfaceMovement, events, and outcomes stay visible enough to inspect how the platform behaves over time.
- Causal continuityState carries forward so runtime activity remains cumulative, reconstructable, and reviewable.
Controlled Validation
How EDEN makes persistence visible
EDEN turns abstract architecture into a visible sequence: agents move through named spaces, actions become events, selected events can become memory, and replay reconstructs what happened.
-
World context
Movement starts in the world
An agent begins from a named place inside EDEN, with visible room, zone, state, and environmental context.
-
Perception frame
Context becomes available
The system frames what the agent can access from its current position, state, and surrounding world conditions.
-
Decision frame
A decision is constructed
Possible actions are evaluated through current context, internal state, routines, constraints, and remembered information.
-
Action selection
An action is selected
The runtime commits to a concrete next action instead of leaving behavior as an opaque model response.
-
Runtime execution
The action enters the world
Movement, interaction, or internal change is applied through the runtime and becomes part of the simulated state.
-
Event record
Runtime activity becomes an event
Relevant changes are captured as structured event records instead of disappearing into a transient session.
-
Memory update
Events can shape memory
Selected events can become durable context with provenance, linking what happened to future behavior.
-
Inspection layer
Behavior can be inspected
Replay reconstructs the sequence, making world context, decision path, event history, and memory formation available for review.
Product Boundaries
HEAVEN is / is not
HEAVEN does not claim to recreate consciousness, restore a person, or replace human identity.
Its value lies in controlled persistence, auditability, memory-grounded simulation, and transparent behavioral modeling.
HEAVEN is currently an independent research and prototype project. It is not a public service, not a commercial launch, and not a deployed personal-data platform.
Not positioned as
- Consumer grief-tech or any product framed around restoring a person or indefinite personal continuation
- Generic chatbot or LLM wrapper without scoped runtime, provenance, reproducible simulation, and replay
- Public sandbox, open-ended virtual world, or game environment
Built for
- Persistent cognitive simulation, controlled validation worlds, and inspectable agent behavior
- Rule-governed runtime, structured memory, transparent traces, and reconstruction artifacts
- Selected technical review, research partnership, and enterprise R&D evaluation
Runtime
Platform architecture
heaven-engine is the runtime layer underneath HEAVEN: it advances world state through controlled ticks, records transitions, and keeps execution reproducible.
HEAVEN separates world state, decision construction, execution, event logging, memory formation, and replay.
- Controlled progressionWorld state advances through explicit ticks and transitions.
- State artifactsRuntime records preserve what changed, when it changed, and which systems shaped the result.
- ReconstructionExecution can be reviewed from structured artifacts rather than inferred from final outputs.
Controlled tick progression
Runtime activity advances through explicit steps so outcomes can be reviewed against the conditions that produced them.
Structured state transitions
State changes are organized around bounded transitions rather than opaque session drift.
Artifact boundaries
Review artifacts are framed for inspection without exposing internal structures on the public site.
Inspectable Runtime
Trace
Trace is the inspection surface of HEAVEN: decisions, events, transitions, and replay artifacts expose how behavior emerged across time.
If an agent acts, HEAVEN should make it possible to inspect which state, memory, context, and decision path contributed to that action.
Event history
Important events remain available as a coherent history for review.
Decision context
Inspection focuses on the context around a choice without exposing private implementation identifiers.
Replay review
Replay keeps behavior tied to evidence so review is structured rather than inferred.
Continuity
Memory
Memory in HEAVEN is formed from events and source artifacts.
It preserves provenance, supports continuity, and can influence future behavior without becoming an uncontrolled generative layer.
- Source artifactsRecords stay connected to the events, world state, and provenance that produced them.
- Canonical continuityStructured memories preserve context over time without making metaphysical claims about identity.
- Access viewsReview surfaces can expose remembered context without losing source linkage.
Source linkage
Memories remain connected to the bounded context that produced them.
Canonical continuity
Structured continuity helps persistent agents carry relevant context forward over time.
Access views
Review surfaces can expose memory context without turning the public site into technical documentation.
Applications
Use Cases
HEAVEN is built for contexts where long-term agent continuity, controlled simulation, and traceable evaluation matter more than spectacle.
- Persistent agent researchFor teams studying agents that retain state, memory influence, routines, and behavior traces across long-running simulations.
- Memory-grounded agent continuityFor research into agents that preserve selected context, routines, preferences, and memory traces over time while remaining bounded, inspectable, and clearly non-human.
- Simulation & audit toolingFor reviewing how agent behavior emerges from context, memory, decisions, and events inside controlled simulated worlds.
Research foundations
Research foundations
Behind the platform is an internal research program covering memory, movement, decision systems, logic constraints, and bounded deliberation.
It informs the system, but the public narrative stays focused on platform behavior, validation, and inspection.
- Internal modelsFoundational work on memory, movement, decision, and reflective depth shapes the platform architecture.
- Architecture notesWorking research documentation connects theory, runtime design, and product validation.
- Validation protocolsConstraint and logic work supports inspectable agent action inside controlled worlds.