The memory layer for long-horizon intelligence
MemoraX is not a product, an agent, or a model.
It is the layer that allows intelligent systems to persist across time.
Modern AI is episodic.
It generates, responds, and resets.
Long-horizon intelligence requires something different:
a layer that carries identity, context, and accumulated understanding forward between interactions.
That layer is memory.
MemoraX is its name.
This repository is the canonical reference for the memory layer of long-horizon intelligence.
It is not an implementation.
It is the definition and namespace of where continuity, identity, and persistence live in the AI stack.
Models reason.
Tools execute.
MemoraX persists.
Every intelligent system is bounded by how far its memory can reach.
MemoraX is the layer that extends that horizon.
These distinctions are non-negotiable.
-
Long context is not continuity
It is just a longer episode. -
Retrieval is not identity
Accessing past information does not preserve self. -
Session memory is not persistence
If restarting the system resets who it is, continuity is absent.
Storing data is not the same as preserving identity.
History alone does not compound understanding.
If the system restarts, is it still the same system?
- If the answer is no → continuity is absent
- If the answer requires explanation → continuity is weak
- If the answer is yes, without qualification → continuity is present
MemoraX defines the canonical memory layer for long-horizon intelligence.
-
Current specification:
SPEC.md(v0.2)
Extended conceptual specification covering scope, lifecycle semantics, governance, and reference use cases. -
Archived specification:
SPEC-v0.1.md
Original foundational definition of identity-preserving memory primitives and core operations.
SPEC v0.1 establishes the core ontology.
SPEC v0.2 extends it with lifecycle and governance semantics while preserving the original primitives and philosophy.