Vercel Adds fx Adapter to AI SDK Harness Layer.

Vercel is bringing its lightweight, open-source coding agent fx into the AI SDK harness layer, giving developers a standard way to run the agent inside applications without maintaining a custom integration.

The company announced the new adapter on August 31, 2026. Developers can now connect fx to the AI SDK through the official @ai-sdk/harness-fx package and control it using the same HarnessAgent interface supported by other coding agents.

That matters because agent integrations can quickly become infrastructure projects of their own. Each coding agent may expose different commands, event formats and connection mechanisms. Vercel’s harness layer puts a common API in front of those differences, making it easier to add or change agents without rewriting the surrounding application.

The basic setup requires importing HarnessAgent and the new fx adapter, then passing fx as the selected harness:

import { HarnessAgent } from '@ai-sdk/harness/agent';
import { fx } from '@ai-sdk/harness-fx';

const agent = new HarnessAgent({
  harness: fx,
});

Once configured, the application runs fx through the HarnessAgent interface. According to Vercel, the adapter communicates with fx over the Agent Client Protocol, or ACP, using the existing @ai-sdk/harness-acp integration.

ACP provides the connective layer between the harness and agent. The dedicated fx package therefore acts as a supported entry point rather than requiring developers to implement the protocol connection themselves. Vercel has published separate fx harness documentation for installation and setup.

The release expands a growing catalogue of supported harnesses. fx joins Claude Code, Cline, Codex, Cursor, Deep Agents, Grok Build, OpenCode and Pi. Having these options behind one interface could be useful for teams evaluating coding agents, building internal developer tools or offering agent-backed features where the underlying harness may change over time.

The immediate benefit is narrower than a new model or autonomous workflow product. Vercel is not announcing additional fx capabilities in this update. It is making the agent easier to embed and operate through an abstraction developers may already use.

For small teams, that can still remove meaningful integration work. Potential use cases include:

• Running fx from an internal development portal
• Connecting coding tasks to a broader agent workflow
• Testing fx alongside other supported coding agents
• Standardising how an application starts and interacts with agent sessions

The shared interface may also reduce switching costs. An application designed around HarnessAgent is less tightly coupled to one agent’s proprietary integration, although differences in agent behaviour and supported capabilities will still need to be handled at the product level.

The addition fits Vercel’s broader effort to assemble infrastructure for agent-based applications. The company has also expanded secure external service access through Vercel Connect reaching general availability, while AI Gateway continues to add model providers and capabilities. Together, those products address separate parts of the stack: harnesses for running agents, Connect for scoped access to outside services and AI Gateway for model access.

For developers already using fx, the new adapter offers a supported route into that stack. For those standardising on the AI SDK harness layer, it adds another coding agent without introducing another standalone integration.

What is Vercel fx?

fx is Vercel’s lightweight, open-source coding agent. The new @ai-sdk/harness-fx adapter allows developers to run it through the AI SDK’s common HarnessAgent interface.

How does fx connect to the AI SDK harness layer?

The official adapter connects to fx over the Agent Client Protocol using @ai-sdk/harness-acp. Developers configure fx as the harness supplied to a HarnessAgent instance.

When is the fx harness adapter available?

Vercel announced that fx was available in the AI SDK harness layer on August 31, 2026. The announcement does not specify a required paid plan or separate adapter price.