The fx adapter now runs the harness on the host
The AI SDK's fx adapter moves the harness and model requests onto the host while keeping built-in shell and filesystem tools sandboxed, so model credentials stay off the sandbox.
The AI SDK's fx adapter now runs the harness and model requests in the **host process**, while built-in filesystem and shell operations remain in the sandbox. It removes the need for an fx install in the sandbox, a bridge, or an exposed sandbox port; model credentials stay on the host.
Builders using HarnessAgent can update `@ai-sdk/harness-fx` to gain tool filtering, mid-turn steering, schema-based structured output, image tool results, and sessions that can resume in a new host process.
The AI SDK's fx adapter now runs the harness and model requests in the **host process**, while built-in filesystem and shell operations remain in the sandbox. It removes the need for an fx install in the sandbox, a bridge, or an exposed sandbox port; model credentials stay on the host. Builders using HarnessAgent can update `@ai-sdk/harness-fx` to gain tool filtering, mid-turn steering, schema-based structured output, image tool results, and sessions that can resume in a new host process. The material does not specify a version number or discuss compatibility and operational tradeoffs of the new host/sandbox boundary.
This clarifies the fx adapter boundary: harness execution, model requests, and credentials stay on the host, while built-in filesystem and shell tools remain sandboxed. It removes the sandbox install and bridge requirements and adds resumable sessions and other HarnessAgent capabilities. The source leaves compatibility and operational tradeoffs unspecified.