AI SDK Vs Little Harness

AI SDK is the closest API-shape influence for Little Harness.

Little Harness should feel like it belongs next to generateText, streamText, ToolLoopAgent, createAgentUIStream, UIMessage[], and useChat. It should not ask developers to abandon the AI SDK chat path.

Use AI SDK Directly When

• the app needs model calls, typed tools, and streamed UI
• message persistence is normal app database work
• tools return compact values that fit in model context
• the agent does not need a managed filesystem
• execution happens safely inside the app's existing server runtime

AI SDK is the right primitive when the application layer is already simple.

Use Little Harness When

• the agent needs a private workspace for scratch files
• tools may return large outputs that should become artifacts
• the agent needs shell/script workflows over files
• skills should be exposed as readable directories
• selected folders should persist through app-owned load/store callbacks
• the host app needs traces that include file changes and persistence failures

Little Harness should use AI SDK conventions while adding the missing harness layer.

Core Difference

How They Should Fit

export async function POST(req: Request) {
  const { messages } = await req.json();

  const result = streamHarness({
    harness,
    messages,
    session: chatId,
  });

  return result.toUIMessageStreamResponse();
}

For non-streaming work, Little Harness should follow the same split as AI SDK:

import { Output } from "ai";

const result = await generateHarness({
  harness,
  type: "account.brief",
  input: { accountId },
  output: Output.object({ schema: accountBriefSchema }),
});

The client stays ordinary AI SDK:

const { messages, sendMessage } = useChat({
  transport: new DefaultChatTransport({ api: "/api/chat" }),
});

Sources

• AI SDK Agents overview
• AI SDK generateText
• AI SDK streamText
• AI SDK ToolLoopAgent
• AI SDK createAgentUIStream
• AI SDK message persistence
• AI SDK telemetry