Ship a run as a product surface

A long piece of work with a progress view is still run(). AgentRun is awaitable and iterable: run.events is the live feed, and awaiting the same object gives you the report at the end.

Vendo’s chat UI is not handed over here, deliberately. This lane gives you an event stream and a report — the surface is yours to build. If you want Vendo’s own thread, overlay, and generated screens, that is the umbrella (createVendo + <VendoProvider>), not this SDK.

The route

// app/api/reports/route.ts
import { THREAD_ID_HEADER } from "@vendoai/agents";
import { support } from "@/lib/agent";
import { auth } from "@/lib/auth";

export async function POST(req: Request) {
  const user = await auth(req);
  if (user === null) return new Response("unauthorized", { status: 401 });

  const { brief } = (await req.json()) as { brief: string };

  const run = support.run(brief, { as: user.id, signal: req.signal });

  const encoder = new TextEncoder();
  const stream = new ReadableStream<Uint8Array>({
    async start(controller) {
      for await (const event of run.events) {
        controller.enqueue(encoder.encode(`data: ${JSON.stringify(event)}\n\n`));
      }
      const report = await run;
      controller.enqueue(
        encoder.encode(`event: report\ndata: ${JSON.stringify(report)}\n\n`),
      );
      controller.close();
    },
  });

  return new Response(stream, {
    headers: {
      "content-type": "text/event-stream",
      "cache-control": "no-cache",
      [THREAD_ID_HEADER]: run.threadId,
    },
  });
}

Three details carry their weight:

run.threadId is available immediately, before the first event — so the header goes out with the response, not after it.
signal: req.signal ties the run to the connection. The browser closes the tab, the request aborts, the run stops. There is no cancel() to call.
await run after the loop is the report. Iterating the events does not consume the result; the same object is both.

What comes over the wire

run.events yields RunEvents, each tagged by type:

`type`	What it marks
`text`	the model’s own words, as they arrive
`status`	where the run is
`tool-call`	a tool the run is about to use
`tool-result`	what that tool answered
`error`	the run failed, with the harness’s own sentence

Forward them whole, as above, and branch in your UI — a new event type then reaches your client the day it ships instead of being swallowed by a server that only knew five.

Ending on the report

The report is the honest close: status, the model’s summary, the toolCalls it made, refs.threadId and refs.approvals, and usage.

// in your client
source.addEventListener("report", (event) => {
  const report = JSON.parse(event.data);
  if (report.status === "stopped") showCancelled();
  else if (report.refs.approvals.length > 0) showApprovals(report.refs.approvals);
  else showDone(report.summary);
});

Anything parked for a human is in refs.approvals. Acting on one of those is a fresh run(), not a resume of this one — see Approvals inside run.

Reference

AgentRun, RunEvent, and the full report shape.

Vendo's own surfaces

The umbrella, if you want the thread, the overlay, and generated screens.

Start here

You already have an agent

Your product has no agent

Agent SDK

Expose your product over MCP

Connect your app

Your brand and your rules

What the agent can do

Go to production

Installing with a coding agent

Reference

Changelog

Ship a run as a product surface

The route

What comes over the wire

Ending on the report

Reference

Vendo's own surfaces

​The route

​What comes over the wire

​Ending on the report

Reference

Vendo's own surfaces

The route

What comes over the wire

Ending on the report