Call built agents from code

LangSmith Product Help
1

Hi everyone,

I’m trying to invoke an Agent Builder agent from my Next.js application using the @langchain/langgraph-sdk package, but my runs fail immediately with status: “error” without any detailed error message.

Setup:

  • Agent Builder agent

  • Model: anthropic:claude-sonnet-4-5

  • Visibility: tenant (workspace-scoped)

  • LangGraph SDK: @langchain/langgraph-sdk (latest)

  • Framework: Next.js 16

What works:

  • :white_check_mark: Authentication with PAT (Personal Access Token)

  • :white_check_mark: client.assistants.get(AGENT_ID) returns the agent successfully

  • :white_check_mark: client.threads.create() works

  • :white_check_mark: client.runs.create() creates a run with status pending

What fails:

  • :cross_mark: The run immediately transitions to status: “error” (within 1 second)

  • :cross_mark: No detailed error message in the run object

  • :cross_mark: client.runs.stream() and client.runs.wait() both throw PermissionDeniedError

My code:

import { Client } from “
/langgraph-sdk”;

const client = new Client({
apiUrl: process.env.LANGGRAPH_API_URL,  // https://prod-deepagents-agent-build-xxx.us.langgraph.app
apiKey: process.env.LANGGRAPH_API_KEY,  // PAT token (lsv2_pt_xxx)
defaultHeaders: {
“X-Auth-Scheme”: “langsmith-api-key”,
},
});

const AGENT_ID = “xxxxx”; //retrieve from code snippet

// This works
const agent = await client.assistants.get(AGENT_ID);
console.log(agent.name); // “Healthcare Appointment Booking Assistant” ✅

// This creates a thread
const thread = await client.threads.create(); // ✅

// This creates a run but it fails immediately
const run = await client.runs.create(thread.thread_id, AGENT_ID, {
input: { messages: [{ role: “user”, content: “Hello” }] },
});
// run.status = “pending” initially, then “error” after 1 second

// This throws PermissionDeniedError
await client.runs.wait(thread.thread_id, AGENT_ID, {
input: { messages: [{ role: “user”, content: “Hello” }] },
});

Environment:

PAT belongs to the agent owner (same ls_user_id)

Agent works fine in the Agent Builder UI

Using the correct API URL from Agent Builder settings

Questions:

  • Why does runs.stream() and runs.wait() throw PermissionDeniedError while runs.create() works?
  • How can I get the actual error details from a failed run?
  • Is there something specific I need to configure for workspace-scoped (tenant) agents?
  • Could this be related to the Anthropic API key configuration in Agent Builder?

Any help would be appreciated! :folded_hands: