This specification extends SPEC v0.1. Core primitives and philosophy remain unchanged. SPEC v0.2 adds scope clarification, lifecycle semantics, governance boundaries, and reference use cases.
1. Purpose
MemoraX defines the memory layer required for long-horizon intelligence.
Modern AI systems are episodic. They act, respond, and reset.
Even with large context windows, retrieval systems, and agent frameworks, most systems fail to remain the same system over time.
Long-horizon intelligence requires:
- persistent identity
- continuity across executions
- compounding understanding
Storage remembers data. MemoraX remembers the system.
2. Position in the AI Stack
- Models → reason
- Tools → act
- Storage → store data
- MemoraX → preserve identity across time
Without this layer, intelligence fragments into isolated episodes.
3. Core Principle
If the system restarts, is it still the same system?
If the answer is “no”, the system lacks long-horizon memory.
MemoraX exists to make the answer “yes”.
4. Scope and Non-Goals
4.1 What MemoraX Is
- an architectural memory layer
- identity-preserving
- continuity-first
- model-agnostic
- compatible with agents, robotics, physical AI, and long-running systems
MemoraX governs what persists and how identity evolves over time.
4.2 What MemoraX Is Not
- a language model
- an agent framework
- a vector database
- a retrieval pipeline
- a chat history store
- a personalization feature
MemoraX does not replace reasoning, retrieval, or storage. It defines the continuity layer between them.
5. Canonical Primitives
5.1 Identity
An Identity represents a single continuous intelligent system.
- persists across executions and deployments
- anchors all memory operations
- remains stable as the system evolves
Identity is not a user. It is the continuous self of an AI system.
5.2 Memory Objects
A memory object contains:
- memory_id
- identity_id
- content
- provenance
- confidence (0.0–1.0)
- created_at
- updated_at
Memory objects are mutable components of an evolving system.
5.3 Continuity Graph
MemoraX organizes memory as a continuity graph.
- what the system knows
- how knowledge was learned
- how understanding changes
- how conflicts resolve
Memory shapes future behavior.
6. Memory Lifecycle Semantics
Context → Candidate → Governance → Committed → Updated/Merged → Deprecated → Archived
- Context — transient
- Candidate Memory — eligible
- Committed Memory — persistent
- Updated Memory
- Merged Memory
- Deprecated Memory
- Archived Memory
This prevents silent identity corruption.
7. Core Operations
- WRITE
- UPDATE
- MERGE
- READ
- DEPRECATE
- AUDIT
Deletion is a governance event.
8. Memory Governance
- what persists
- how memory evolves
- how drift is detected
- how accountability is enforced
Without governance, persistence leads to drift.
9. Relationship to Retrieval and Storage
Retrieval is episodic. MemoraX is continuous.
Retrieval answers questions. MemoraX shapes behavior.
10. Reference Use Cases
- Long-running agents
- Persistent support systems
- Robotics
- Long-term human–AI collaboration
11. Relationship to Implementations
MemoraX defines the abstraction layer. Implementations may vary.
12. Status
Conceptual. Vendor-neutral. Implementation-agnostic.
It defines the memory layer required for identity-preserving long-horizon systems.
Closing
The future remembers.
Intelligence compounds.
It starts with MemoraX.