Planner not working in production (working in local development)

[phy101]

Getting this error in langgraph cloud

The manager works but fail when call the planner.

HTTP 404: Not Found

Error: HTTP 404: Not Found
at HTTPError.fromResponse (file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:56:20)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:132:23
at async RetryOperation._fn (/deps/open-swe/node_modules/p-retry/index.js:50:12)

[fernando-plank]

I'm getting this error too, but after making the changes to the bodyParams, I was previously unable to reach this stage due to the config attributes. looking at the cloud metadata, I'm missing the access token field.

[phy101]

yes, I fixed the body params ( #818 ). But now I'm stuck on this 404 error in the planner. It it likely to be related to the proxy but still debugging.

[fernando-plank]

In my case is not totally fixed. If I try after the error interact with the chat will fail and in the metadata fields on langgraph cloud will miss some fields like installation name and the token.

[phy101]

The code itself looks good, I revised several time and it runs locally without error. I wonder if is any langgraph configuration that I'm missing.
the deploy-langgraph.yml asks for those keys in github actions. I couldn't find the integration_id. But I have no idea if this is the cause of the error.

@bracesproul could be kind to clarify if those need to be set to deploy in langgpraph cloud?

    env:
      CONTROL_PLANE_HOST: ${{ secrets.CONTROL_PLANE_HOST }}
      LANGSMITH_API_KEY: ${{ secrets.LANGSMITH_API_KEY }}
      INTEGRATION_ID: ${{ secrets.INTEGRATION_ID }}
      DEPLOYMENT_ID: ${{ secrets.DEPLOYMENT_ID }}
    timeout-minutes: 35

[phy101]

I'm pretty sure is some langpraph cloud configuration that I'm missing. I revised the code extensively and couldn't find any issue that could cause the 404 error. The langgraph.json was updated today, it solves the x-github-installation-id issue, but I still getting the 404 error in production.

[phy101]

Even with the langgraph.jsson update I still getting the 404 error in production (work in local dev) when start the planner, the manager works.

ehre the full server log in langgraph cloud

[ERROR] Background run failed. Exception: <class 'langgraph_api.js.errors.RemoteException'>(HTTP 404: Not Found)
9/4/2025, 9:39:23 AM
[ERROR] Run encountered an error in graph: <class 'langgraph_api.js.errors.RemoteException'>(HTTP 404: Not Found)
Traceback (most recent call last):
File "/api/langgraph_api/worker.py", line 142, in wrap_user_errors
File "/api/langgraph_api/stream.py", line 462, in consume
File "/api/langgraph_api/stream.py", line 445, in consume
File "/api/langgraph_api/stream.py", line 246, in astream_state
File "/api/langgraph_api/asyncio.py", line 89, in wait_if_not_done
File "/api/langgraph_api/js/remote.py", line 154, in astream_events
File "/api/langgraph_api/js/remote.py", line 104, in _client_stream
langgraph_api.js.errors.RemoteException: HTTP 404: Not Found
9/4/2025, 9:39:23 AM
[ERROR] Error: HTTP 404: Not Found
at HTTPError.fromResponse (file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:56:20)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:132:23
at async RetryOperation._fn (/deps/open-swe/node_modules/p-retry/index.js:50:12)
9/4/2025, 9:39:23 AM
[INFO] }
9/4/2025, 9:39:23 AM
[INFO] run_id: '019914f3-e08c-74c8-8f13-27fdc7d84319'
9/4/2025, 9:39:23 AM
[INFO] thread_id: 'e85e0e3e-3f46-4f96-9270-d17b3c8c0d29',
9/4/2025, 9:39:23 AM
[INFO] ' at async RetryOperation._fn (/deps/open-swe/node_modules/p-retry/index.js:50:12)',
9/4/2025, 9:39:23 AM
[INFO] ' at async file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:132:23\n' +
9/4/2025, 9:39:23 AM
[INFO] ' at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n' +
9/4/2025, 9:39:23 AM
[INFO] ' at HTTPError.fromResponse (file:///deps/open-swe/node_modules/@langchain/langgraph-sdk/dist/utils/async_caller.js:56:20)\n' +
9/4/2025, 9:39:23 AM
[INFO] stack: 'Error: HTTP 404: Not Found\n' +
9/4/2025, 9:39:23 AM
[INFO] message: 'HTTP 404: Not Found',
9/4/2025, 9:39:23 AM
[INFO] name: 'Error',
9/4/2025, 9:39:23 AM
[INFO] [StartPlanner] Failed to start planner {
9/4/2025, 9:39:23 AM
[DEBUG] Incremented nodes executed to 2
9/4/2025, 9:39:23 AM
[INFO] GET /threads/e85e0e3e-3f46-4f96-9270-d17b3c8c0d29 200 144ms

[fernando-plank]

I will try again on my side and let you know, thanks for keeping me on track too @phy101!

[phy101]

keep me posted if you figure it out. I will wait for a solution, it is not a code a code bug, but a configuration mismatch and the Langraph cloud docs didn't help.

[n-sviridenko]

My assumption is that the integration id is the deployment id. But I don't see it being used so prob skipping is fine. @phy101

[n-sviridenko]

Actually, I have the same issue. And I think it's due to a missing assistant. Somehow the system doesn't create a default assistant when provisioning.

[n-sviridenko]

And I think it's due to the authentication settings that probably don't let the assistant to be created (?)

Due to that extra github auth stuff.

[phy101]

Since it works fine in local development it is likely to be some langgraph cloud misconfiguration or security settings that I couldn't identify..

It is related to the authentication: The proxy route ensures that sensitive authentication tokens never reach the client directly, maintaining security while enabling seamless communication with the LangGraph server.

https://docs.langchain.com/labs/swe/setup/authentication#langgraph-server-authentication

[n-sviridenko]

@phy101 I found the root cause:

open-swe/apps/open-swe/src/utils/langgraph-client.ts

Lines 8 to 20 in 19debf1

	const productionUrl = process.env.LANGGRAPH_PROD_URL;
	const port = process.env.PORT ?? "2024";
	if (options?.includeApiKey && !process.env.LANGGRAPH_API_KEY) {
	throw new Error("LANGGRAPH_API_KEY not found");
	}
	return new Client({
	...(options?.includeApiKey && {
	apiKey: process.env.LANGGRAPH_API_KEY,
	}),
	apiUrl: productionUrl ?? `http://localhost:${port}`,
	defaultHeaders: options?.defaultHeaders,
	});
	}

^-- this env was not set and it defaulted to localhost

[n-sviridenko]

cc @fernando-plank

[phy101]

great, I will test it.