| name | shared-state-architecture |
| description | PSI (Persistent Shared Interface): A shared-state architecture for coherent AI-generated instruments in personal AI agents. Addresses the problem of isolated AI tools by introducing a personal-context bus for cross-module reasoning and synchronized actions. Use when designing multi-agent systems, personal AI environments, tool orchestration, or building coherent AI software architectures. |
PSI: Shared State Architecture for Personal AI Agents
Core Problem
AI-generated personal tools remain isolated after creation, lacking coordination mechanisms. Each tool operates independently without shared context, leading to fragmented user experiences.
Key Innovation: Personal-Context Bus
PSI introduces a shared-state layer that transforms isolated AI modules into coherent instruments through:
- State Publication: Modules publish current state to shared bus
- Write-back Affordances: Modules expose editable state properties
- Cross-module Reasoning: Chat agent can reason across module states
- Synchronized Actions: Multiple modules can coordinate through shared state
Architecture Components
1. Personal-Context Bus
- Central communication channel
- Publish/subscribe pattern for state updates
- Supports both read and write operations
2. State Contract
class ModuleState:
module_id: str
state_type: str
current_state: dict
writeable_properties: list[str]
def publish_state() -> None
def subscribe_to_state(module_id: str) -> None
def write_state(property: str, value: Any) -> bool
3. Integration Pattern
- New modules automatically inherit shared-state contract
- Existing modules can be wrapped with state adapters
- Chat agent accesses all module states uniformly
Design Principles
- Persistence: State survives across sessions
- Coherence: All modules see consistent view
- Chat-complementary: Both GUI and chat interfaces work
- Automatic integration: New tools join ecosystem seamlessly
Use Cases
- Personal AI assistants with multiple tools
- Cross-tool workflow automation
- State-aware AI agents
- Multi-modal personal computing environments
Implementation Considerations
- State conflict resolution (CRDTs, versioning)
- Privacy boundaries between modules
- Performance under high update frequency
- Backward compatibility with existing tools
Paper Reference
Title: PSI: Shared State as the Missing Layer for Coherent AI-Generated Instruments in Personal AI Agents
Authors: Zhiyuan Wang, Erzhen Hu, Mark Rucker, Laura E. Barnes
arXiv: 2604.08529
Category: cs.HC, cs.AI
Published: 2026-04-09
Related Concepts
- Shared memory in multi-agent systems
- Event-driven architecture
- Actor model for state management
- Context-aware computing