Agentic State Sovereignty
State belongs to the Protocol. Execution belongs to the Runtime.
The Problem
Who Owns an Agent’s State?
In today’s agent frameworks, state is implicitly owned by the runtime.
Agent memory, context, task history, and execution state are often:
- Encoded in runtime-specific data structures
- Persisted in proprietary formats
- Tightly coupled to a specific orchestration engine
As a result:
- Agents are not portable
- Memory cannot be reliably migrated across runtimes
- Vendor lock-in becomes structural, not incidental
This creates a fundamental architectural flaw: the “identity” of an agent is inseparable from the tool that executes it.
Core Principle
State Belongs to the Protocol
MPLP adopts a different architectural stance.
“State belongs to the Protocol. Execution belongs to the Runtime.”
Semantic Definition
The meaning and structure of agent state are defined at the protocol layer.
Execution Scope
The runtime is responsible only for execution, scheduling, and resource management.
No Ownership
No runtime is considered the canonical owner of agent memory.
In MPLP, state is a semantic construct, not an implementation artifact.
Architecture
Protocol-Level State Definition
MPLP introduces protocol-defined structures that describe agent state in a runtime-neutral way.
Protocol Structures
- •Context frames
- •Lifecycle stages
- •Semantic relationships between tasks, decisions, and outcomes
What These Define
- ✓What constitutes valid state
- ✓How state evolves across the agent lifecycle
- ✓What information must be preserved for continuity and auditability
Runtimes may interpret or store this state differently, but they do not redefine its meaning.
Roadmap
Portability as an Architectural Goal
Portability is treated as a design objective, not a v1.0 promise.
Agentic State Sovereignty does NOT claim that:
- ✗All runtimes can interchange state seamlessly
- ✗Any existing implementation is fully portable today
- ✗Migration is lossless or automatic
Instead, MPLP establishes:
- ✓A protocol-level boundary for state semantics
- ✓A shared reference model that runtimes may choose to support
- ✓A foundation upon which portability can be incrementally achieved
Impact
Why This Matters
By separating state from execution, MPLP enables a healthier ecosystem.
Long-lived Agents
Agents that outlast specific tools or framework versions.
Auditable Memory
Memory that is not locked to a vendor's proprietary format.
Fair Competition
Future ecosystems where runtimes compete on execution quality, not data captivity.
“Agents should not lose their memory when their runtime changes.”
MPLP defines agent state as a protocol concern, leaving execution strategies to interchangeable runtimes.