The architecture for systems that persist.
Neumera is the architecture for systems that persist: systems that carry memory, commitments, and identity through time while staying attached to the same world.
Softminds are not chat turns with a memory buffer. They are continuous entities that maintain internal structure, remain world-participating, and stay governable while their behaviour evolves.
How Neumera behaves
Continuity without world attachment drifts. World attachment without governance becomes unsafe. Governance without evidence becomes opaque. Evidence without continuity cannot explain an evolving entity. The system works because these conditions are coupled.
- continuity: the same live system persists across time
- persistent internal structure: commitments, unresolved state, and role remain live
- worlds: shared objects, constraints, and consequences endure
- governance: permissions, intervention, and constitutional limits remain attached
- evidence: history is inspectable, replayable, and reviewable
How Softminds navigate the world
A softmind is not an isolated prompt loop. It remains coupled to world objects, tasks, and changing conditions. That coupling is what makes responsibility, repair, and long-term coordination possible.
What softminds are not: they are not agent scripts, not stateless orchestration wrappers, and not one-shot automation pipelines with a memory addon.
System overview
Neumera is not one mechanism. It is a stack of conditions that have to exist together if a system is to remain continuous, world-participating, governable, and inspectable across time.
Runtime
Runtime sustains continuity, internal transformation, active commitments, and the carrying forward of unresolved state. This is where the system remains the same system across time instead of becoming a fresh event at every turn.
World
World provides persistent objects, conditions, history, and constraints. It is what lets a system remain attached to the same task, the same object, and the same consequences rather than drifting across disconnected prompts.
Governance
Governance defines permissions, intervention, limits, and constitutional structure. A system that persists cannot be governed as an afterthought. Control has to remain attached to the same live entity and the same evolving context.
Evidence
Evidence preserves traces, replay, reports, and inspectable history. If a system changes over time, we need to see what changed, why it changed, and what remained continuous.
These layers do not sit beside one another as optional features. They create the minimum conditions under which a softmind can persist as a real entity in a real world.
Reasoning, interaction, coordination, and learning.
Softminds are fundamentally different to LLMs, reinforcement learning and state machines. Read our FAQ
How can Softminds reason without training data?
Softminds generate behaviour from their current state rather than from pre-learned examples.
At any point in time, the system carries:
- active tasks and commitments
- structured world state
- constraints and conditions
- accumulated history
Reasoning emerges from evaluating how possible actions change that state.
Each decision is produced by:
- what is currently true
- what is required
- what is allowed
- what follows from each option
This produces coherent behaviour without relying on stored patterns of prior responses.
How do Softminds understand the world without priors?
Softminds build understanding from the world they are actively part of.
They operate on:
- explicit objects
- defined tasks
- observed events
- shared contexts
Meaning comes from:
- what has been encountered
- what has changed
- what remains active
- what is at stake
This produces a situated understanding of the world that develops through interaction and continuity.
How do Softminds evaluate different paths?
Softminds evaluate actions in terms of their effects on the current world state.
For each possible path, the system considers:
- how task status changes
- whether constraints are satisfied
- how commitments are affected
- how continuity is preserved
This evaluation is grounded in the live system rather than in similarity to past examples.
Alternative paths are compared within the same state, allowing decisions to reflect real tradeoffs.
How do Softminds operate in an open and changing world?
Softminds operate with the expectation that the world is incomplete and continuously changing.
External changes enter as events that:
- modify world state
- introduce new constraints
- invalidate prior assumptions
The system incorporates these events and updates its behaviour while maintaining the same responsibility.
This allows continuity even when conditions change in ways that could not be anticipated in advance.
How do Softminds handle the boundary problem?
Softminds operate within a defined world while remaining responsive to external influence.
They:
- act on what is currently known
- maintain explicit links between actions and assumptions
- detect when those assumptions no longer hold
When new information arrives, its impact is introduced into the system as structured change.
This allows the system to adapt continuously while remaining grounded in its current responsibilities.
How do Softminds use language and external data?
Softminds interact with external information through a membrane layer.
The membrane translates between:
- unstructured inputs such as text, documents, images, and video
- structured internal state
Language models can be used inside this layer to interpret and render information.
The core system continues to operate on structured state, while language and media remain an interface to the outside world.
How do Softminds coordinate with each other?
Softminds coordinate through shared contexts.
Within a shared context, they maintain:
- commitments
- relationships
- evolving representations of other participants
- active areas of agreement and disagreement
Coordination emerges from:
- shared state
- shared responsibility
- continuity of interaction over time
This allows alignment, conflict, and repair to remain part of one ongoing system.
How does the system remain inspectable?
Every step in a Softmind is recorded as structured change.
The system produces:
- chronological state transitions
- causal links between actions and outcomes
- replayable execution paths
This makes it possible to:
- trace how decisions were formed
- understand how they evolved
- verify behaviour against recorded state
Inspection is part of the system itself, not a layer added afterwards.
Softminds operate as continuous systems where reasoning, adaptation, and coordination arise from evolving state rather than pre-learned responses.