# AI Agent Portability The ability to transfer an [[AI Agents|AI agent]] across platforms, machines, teams, organizations, and model providers while preserving its behavior, identity, and capabilities. Harder than [[AI Skill Portability]] because agents bundle multiple layers: identity, memory, skills, routing, and harness configuration. An agent you build should work on all your machines, for every developer on your team, across your organization, and ideally be shareable with anyone. It should also survive switching from one AI model or provider to another. See [[AI Interoperability]] for the broader principle. ## What's portable vs what's not | Layer | Portable? | Why | |-------|-----------|-----| | **[[AI Agent Identity]]** | Mostly yes | Role definitions, personality, expertise are plain text | | **[[AI Agent Skills]]** | Partially | Depends on [[AI Skill Portability]]; tool calls are platform-specific | | **[[AI Agent Memory]]** | Partially | Memory format varies; content is portable, storage mechanism isn't | | **[[AI Agent Routing]]** | No | Routing infrastructure is tool-specific | | **[[AI Agent Harness]]** | No | Harness is deeply tied to the runtime (Claude Code, Cursor, etc.) | The core insight: **identity is portable, runtime is not**. An agent's personality, expertise, and behavioral rules transfer easily. The infrastructure that runs the agent doesn't. ## Why it matters - Avoids platform lock-in for agent investments - Cross-machine: your agents work identically on every machine you own - Cross-team: agents shared across a team work regardless of individual setups - Cross-org: agents deployed at [[Enterprise Context Management (ECM)|enterprise scale]] work for everyone - Model/provider freedom: switching AI models or providers doesn't break agents - Enables multi-tool strategies (different tools for different tasks, same agents) - Supports [[AI Agent Distribution]] across teams using different tools ## Current state - Agent identity (SOUL.md) is plain Markdown; transfers anywhere - Skill manifests need adaptation per tool - Memory systems are tool-specific (Claude Code's MEMORY.md vs Cursor's approach) - No cross-platform agent runtime standard exists yet - [[Agent Client Protocol (ACP)]] is an emerging attempt at cross-platform agent communication ## References - ## Related - [[AI Agents]] - [[AI Skill Portability]] - [[AI Skill Portability Checklist]] - [[AI Interoperability]] - [[AI Skill Resilience]] - [[AI Agent Distribution]] - [[AI Agent Identity]] - [[AI Agent Memory]] - [[AI Agent Skills]] - [[AI Agent Harness]] - [[Agent Client Protocol (ACP)]] - [[Claude Code]] - [[Cursor.com]]