404 on /api/langgraph/threads/{threadId}/commands when running assistant-ui-claude example

OSS Product Help
1

Hi everyone,

I’m trying to run the examples/assistant-ui-claude example from the LangGraphJS repository.

After some setup issues (workspace dependencies, CLI naming, etc.), the application now starts successfully:

  • Frontend is running on http://localhost:5174
  • LangGraph server is running on http://localhost:2024
  • The UI loads correctly

However, when I try to interact with the assistant, I get a 404 error on the following endpoint:

http://localhost:5174/api/langgraph/threads/<thread-id>/commands

Example:

/api/langgraph/threads/019ebc56-cd63-7742-b9a5-6d4e65308db9/commands

The request returns:

404 Not Found

A few questions:

  1. Is the /commands endpoint still expected to exist in the current version of the example?
  2. Has the Assistant UI example become incompatible with recent LangGraph/LangGraph SDK versions?
  3. Is there an additional proxy or API route configuration that needs to be enabled?
  4. Has anyone successfully run the assistant-ui-claude example recently from the current main branch?

Any guidance would be appreciated.

Thanks!

2

hi @gdrouet

I haven’t tried it lately, but from code investigation I’ve just proceeded I see this:

Root cause

src/components/my-assistant.tsx:25:

const apiUrl =
  import.meta.env.VITE_LANGGRAPH_API_URL ??
  `${window.location.origin}/api/langgraph`;   // -> http://localhost:5174/api/langgraph

No VITE_LANGGRAPH_API_URL set + vite.config.ts has no proxy → request never reaches :2024. Thread id shows up because new SDK transport creates thread client-side before first command POST.

Per question

  1. /commands exist? Yes. langgraph-api v1.3.0 registers it - src/api/protocol.mts:137 api.post("/threads/:thread_id/commands", …). New thread-centric HTTP+SSE transport (libs/sdk/.../transport/http.ts: commandsUrl = /threads/${threadId}/commands)
  2. Incompatible with recent SDK? No. Example + react SDK + langgraph + langgraph-api all same monorepo, pinned together (workspace:*). Config bug, not skew
  3. Extra proxy/route config needed? Yes - the fix
  4. Runs from main? Yes, once configured

Fix - pick one

A. Point SDK at dev server. Make examples/assistant-ui-claude/.env:

OPENAI_API_KEY=your_openai_api_key
OPENAI_MODEL=gpt-4.1-mini
VITE_LANGGRAPH_API_URL=http://localhost:2024
VITE_LANGGRAPH_ASSISTANT_ID=assistant-ui-claude

B. Same-origin + Vite proxy in vite.config.ts:

server: {
  proxy: {
    "/api/langgraph": {
      target: "http://localhost:2024",
      changeOrigin: true,
      rewrite: (p) => p.replace(/^\/api\/langgraph/, ""),
    },
  },
},

(leave VITE_LANGGRAPH_API_URL unset).

Gotcha - second bug after URL fix

Code default assistantId: "claude" ≠ langgraph.json graph key assistant-ui-claude. Must set VITE_LANGGRAPH_ASSISTANT_ID=assistant-ui-claude or next error = graph-not-found.

Restart dev after .env edit - Vite reads .env from example folder, exposes only VITE_-prefixed vars.

===

Could you try it out and let me know whether it works for you?

3

Thanks it works now.

I’m testing with my own agent server built with the docker image langchain/langgraph-api:3.13and it seems /commands does not exist. Am I missing something?

4

hmmmm

can you try one of these?

  1. Upgrade server to v3-capable langgraph-api (bump the server package, not the Python tag), rebuild image, re-probe. Keeps assistant-ui wiring unchanged.

  2. Use the classic Runs hook if you can’t move the server:

import { useStream } from "@langchain/langgraph-sdk/react"; // v1.9.x — hits /threads/{id}/runs/stream

@langchain/react’s useStream can’t downgrade to the Runs API. The classic hook (stream.lgp.tsx → client.runs.stream) works against current Platform servers