Not able to schedule Crons using Langgraph SDK

LangSmith Product Help Deployment
1

I am trying to use Langgraph SDK and using client.crons.create() method. It asks me to pass assistantId where I pass my graph name “agent”, as shown in docs Use cron jobs - Docs by LangChain . But it throws an error that it expects UUID. I assume the default assistant is created based on graph, but when I try to get the list of assistants, it returns empty .

Also when creating stateless jobs, I don’t want to pass threadId, but it still forces me to do so. My purpose is to create new thread on every cron run.

2

Hey! Thanks for raising these concerns and sorry for the pain here. I’m looking into these issues and will let you know when resolved.

To confirm, you’re hitting a postgres server (not in memory) when running these commands correct? Crons aren’t yet supported on the in memory implementation and might be more out of date as a result with some of the validation issues.

3

Thanks Josh. I have been trying on local (in-memory). So if it is not available on local, how to test things? Any other ways?

Also can you confirm why I am getting empty assistants in in-memory? Do we really need to pass UUID or graph name should be enough?

One more thing, I tried to test it on deployed version, and it is able to create cron. But I don’t see it running anywhere. Is there a way at least where I can check if it ran or not or if there are any errors. In server logs I can just see it created the cron successfully, but where to check whether it has triggered or not?

4

Updates:

  1. I am able to schedue cron using deployed version
  2. I was trying to schedule it in non-UTC time so I could not see it run. After fixing this, I can see it gets triggered.

But now the issue I am facing is that when it starts running, I see below error:

28/11/2025, 15:10:04

[ERROR] Error scheduling cron run cron_id=019ac9d3-a644-7719-b577-2b89264cb4e8
Traceback (most recent call last):
File “/api/langgraph_api/cron_scheduler.py”, line 38, in cron_scheduler
File “/api/langgraph_api/models/run.py”, line 305, in create_valid_run
starlette.exceptions.HTTPException: 404: Thread or assistant not found.

Below is the value I see in authorization handler:

[DEBUG] [LangGraph Auth] Value: {
“thread_id”: “955779ac-66b8-4a40-81e0-90f54ea2dfe2”,
“assistant_id”: “a2193a9d-c81f-5f9e-8fb5-76310a5fd9e5”,
“run_id”: “019ac9d5-83f4-733c-8b71-0413ca0f1c40”,
“status”: “pending”,
“metadata”: {
“userId”: “”,
“accountId”: “”,
“scheduledTaskTitle”: “list in progress work orders”,
“assistant_id”: “a2193a9d-c81f-5f9e-8fb5-76310a5fd9e5”
},
“prevent_insert_if_inflight”: false,
“multitask_strategy”: “enqueue”,
“if_not_exists”: “reject”,
“after_seconds”: 0,
“kwargs”: {
“input”: {
“messages”: [
{
“role”: “user”,
“content”: “in progress work orders”
}
]
},
“command”: null,
“config”: {
“configurable”: {
“cron_id”: “019ac9d3-a644-7719-b577-2b89264cb4e8”,
“langgraph_auth_user”: {
“identity”: “Kt8AR2w8ls”,
“is_authenticated”: true,
“display_name”: “Kt8AR2w8ls”
},
“langgraph_auth_user_id”: “Kt8AR2w8ls”,
“langgraph_auth_permissions”: ,
“langgraph_request_id”: null,
“_after_seconds_”: 0
},
“recursion_limit”: 200
},
“context”: {
“cron_id”: “019ac9d3-a644-7719-b577-2b89264cb4e8”
},
“stream_mode”: [
“values”
],
“interrupt_before”: null,
“interrupt_after”: null,
“webhook”: null,
“feedback_keys”: null,
“temporary”: false,
“subgraphs”: false,
“resumable”: false,
“checkpoint_during”: true,
“durability”: “async”
}
}

5

Hey! Thanks for the updates and glad you were able to get them created. To your questions:

So if it is not available on local, how to test things? Any other ways?

You should be able to run npx /langgraph-cli up to start up the postgres server locally, however trying that myself, I’m also running into issues with crons. Apologies and working on a fix for that.

Also can you confirm why I am getting empty assistants in in-memory? Do we really need to pass UUID or graph name should be enough?

Graph name should be fine. The validation in the in-memory version for crons is just out of date. Are you running into issues when calling the assistants search API ( LangGraph Platform API Reference )? Just tried that myself with the graph_id parameter and was able to find a local graph. Can you share structure of the call you’re making there?

starlette.exceptions.HTTPException: 404: Thread or assistant not found.

Can you share the request that you’re making to create cron ( LangGraph Platform API Reference )? We just added support (apologies not yet documented) for the ability to create a new persisted thread for each cron run with this endpoint by passing the parameter on_run_completed set to keep. For example:

curl https://langchain-ai.github.io/runs/crons 
  –request POST 
  –header 'Content-Type: application/json'
  –data '{
    "schedule": "*/5 * * * *",
    "assistant_id": "agent",
    "on_run_completed": "keep"
}'

Hopefully that helps and hope to have fixes for the other issues raised by next week.

6

Hi Josh,

Thanks for the quick reply. It really means a lot as it blocks my prod deployment.

Right now I am able to create stateless crons. They do get executed and I can see the run in Langsmith. Interesting thing is how to access the thread it created (in stateless). I can see the thread in Langsmith but if I try to access it in my application, it says not found.

Is this related to “on_run_completed”: “keep” ?

By the way I am using JS SDK (not python), and in the SDK, for the onCompletion, I see only 2 options: complete and continue. There is no “keep” option.

image
image874×424 29.1 KB

Also could you please help me understand one thing:

I want to create “new” thread for “each” run, so is creating stateless crons the option? Because if I use stateful cron, it uses single thread for each execution, which is not what I want. Reading the docs about stateless, I am not that sure what fulfills my scenario. Ultimately I want each new thread to be created on each run and user should be able to access that thread to resume conversation from there.

Also because there is no documentation, I would like you to confirm below:

I have custom auth implemented. So I have authenticate method and on handler. If I start thread manually it hits both authenticate and on handler, but if cron executes the run, it bypasses authenticate method. It hits on handler though. Is this the way it should work?

7

Just to add more, I also tried calling the API directly as you suggested where I have set on_run_completed: ‘keep’. I can see the thread is created by the cron and visible in Langsmith. But I can’t see the thread tied to the user. How does this work behind the scenes?

image
image1209×1085 148 KB