Vercel's fx adapter now runs the harness on the host process
The fx adapter for the AI SDK's HarnessAgent now runs the harness and model requests in the same process as the application, using the libfx TypeScript library, while built-in filesystem and shell operations remain isolated in the sandbox. The adapter no longer installs fx inside the sandbox, runs a sandbox bridge, or exposes a sandbox port, and model credentials stay on the host instead of being forwarded into the sandbox.
Why it matters
Builders no longer need sandbox-side fx installation, a bridge process, or an exposed port to run agents, and model API keys are not passed into the sandbox. Sessions can now be saved and resumed in a new host process, and the rewrite adds built-in tool filtering, mid-turn steering, structured output, and image tool results via toModelOutput().
Who should care
Developers building AI agents with the Vercel AI SDK and HarnessAgent, especially teams handling model credentials in sandboxed agent workloads.
What you can do
Update @ai-sdk/harness-fx to the latest version and review the fx harness documentation to migrate to the new host-side adapter.