Adapter boundary
The adapter boundary keeps runtime-specific execution, memory, sessions, and media local while making the runtime visible enough to operate from the OpenOrange dashboard.
// attach.runtime_adapter
Runtime adapters for AI agents
OpenOrange runtime adapters connect OpenClaw, Hermes, IronClaw, ACP, webhook agents, channels, tools, and future execution layers to the same operator contract.
Runtime adapters let OpenOrange govern different AI execution shapes without pretending they are all the same runtime. Each adapter exposes the signals, routes, actions, and proof the operator layer needs.
01
Adapters keep execution details in the runtime.
02
The operator layer receives normalized health, request, usage, and audit signals.
03
OpenClaw is one supported runtime shape, not the whole product boundary.
04
Future runtimes can plug into the same plan-first operating contract.
Supported shapes
The model fits OpenClaw, Hermes, IronClaw, ACP, webhook agents, channel agents, custom tools, and future execution layers that can expose health, events, actions, and usage.
Why adapters matter
AI infrastructure changes quickly. Adapters let the operator contract stay stable while runtime implementations evolve underneath it.
// questions
Is OpenOrange only OpenClaw-backed?
No. OpenClaw is one runtime shape. OpenOrange is broader: an operator layer for multiple runtimes, adapters, model routes, tools, and channels.
Do adapters copy runtime state into OpenOrange?
No by default. They expose operated metadata, references, health, events, and actions while live runtime state stays where it runs.