Vercel's fx coding agent joins the AI SDK harness layer via official adapter

fx lands in the AI SDK harness layer

Vercel has added fx, its compact open-source coding agent, to the AI SDK's harness layer. According to a changelog entry on the company's blog dated 31 August 2026, applications that already use the harness layer can now run fx through the same standardized interface they use for other agents, without writing a bespoke integration to support it.

The harness layer addresses a familiar pain point for developers embedding agents in their products: every coding agent tends to arrive with its own setup steps, runtime assumptions and API surface. By putting agents behind a common HarnessAgent abstraction, the AI SDK lets an application interact with any supported agent in a uniform way.

Wiring it up

The integration is handled by an official adapter package, @ai-sdk/harness-fx, and requires only a few lines of configuration:

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

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

Once the HarnessAgent is constructed with fx as its harness, the agent is driven entirely through the HarnessAgent interface, Vercel explains. Under the hood, the adapter communicates with fx over the Agent Client Protocol (ACP), building on the existing @ai-sdk/harness-acp package for the protocol transport.

A growing roster of supported agents

With this release, fx becomes the newest addition to the harness layer's lineup. It sits alongside Claude Code, Cline, Codex, Cursor, Deep Agents, Grok Build, OpenCode and Pi, bringing the number of supported harnesses to nine. For developers already standardised on the HarnessAgent interface, the practical effect is that fx becomes a configuration choice rather than an integration project.

Vercel points readers to the fx harness documentation for getting started with the adapter.

Why it matters

The release lowers the barrier to adopting fx in production applications. Teams using the AI SDK harness layer can add or swap agents with a configuration change, which makes it far easier to evaluate fx against alternatives like Claude Code, Codex or OpenCode and to switch if one agent proves a better fit for a given workflow.

The choice of ACP as the transport is also notable. Because the adapter speaks an open protocol rather than a Vercel-specific wire format, fx interoperates through the same channel used by other ACP-compatible tooling, keeping the integration path consistent with the broader ecosystem of agent clients and harnesses.

Finally, fx's position as a lightweight, open-source agent makes it a low-friction entry point for developers who want to experiment with coding agents inside their own applications. Bundling it into the AI SDK's unified harness API means that experimentation no longer requires building and maintaining separate plumbing for each agent tried along the way.