Constantly getting error in API Request Failed #571
Comments
|
I have this same issue with claude sonnet github version |
|
has similar bug, set the Deepseek API two days ago and it went well, but after the update, I always stuck at the stage of waiting for API response. |
|
Actually, the bug is that when we enter the model name, it doesnt accept capital letters and changes to small letters automatically. This causes many api providers, who use capital letters in their name to fail. |
I encoutered the same problem. If you have any solution, please share with me, thanks. |
|
Oh I know we fixed that lowercase issue someplace last week - will try to track it down and see if we need to fix in more places. |
|
DeepSeek was having API issues this morning btw, in case that’s related to any of the conversation in this thread |
|
I'm having the same issue with the DeepSeek models |
|
Same issue; requests to Deepseek are working as I can do so directly from my scripts. Roo Code however, not working. |
|
Same .... and when trying to utilize R1 I get the following almost every time ...
|
|
For the lowercase issue, can you make sure you're on the latest version? We released what I think should have fixed this in #537 |
|
Roo Code and Cline are not working with R1/V3 for me — they get stuck on API requests with no response, eventually timing out. Cursor works fine with R1. Could this be a block on DeepSeek’s side? |
Now that's interesting. I hadn't thought of that. Looks like I'll be requesting a refund with lawyers if that turns out to be the case. It was so good. I know they have had API issues today but at the time of my issues, it was apparently 100% up with no problems. |
Actually, I just tried curl and some other apps — they’re also getting stuck and timing out. So, most probably, they are experiencing some issues with infrastructure and this is not a block of AI code gens. |
That's exceptional news. I guess I'll just wait then. Deepseek V3 absolutely kills it in this extension. It would be incredibly dumb to revoke access their end in my opinion (the cache-hitting is just chef's kiss) |
|
I think Deepseek's API has been experiencing issues over the past couple of days, and it should be in the process of being restored now.
|
Oh, I see. I’ve been checking the status over the past couple of hours, and it was showing green just an hour ago. That explains everything. Hopefully, they fix it soon so I can get back to my cheap coding 😂 |
|
I should've included in the issue description, but my API provider is Deepinfra, so we can't consider this issue as only happening by contacting directly Deepseek's servers. |
|
Well the status now says that it's fixed and they are monitoring the situation but I'm still not able to get a single call to the API to go through. I'm very concerned at this point. |
Surely they just route through Deepseek anyway? |
This makes me believe it's not on the API providers' side - my guess is that there's some code at the client side (Roo Code) parsing the response wrongly. I'd guess it's a very simple change when dealing with the response, either streamed or not. Nothing to be concerned about :) |
Pretty sure they serve the models in their own infra. |
Fair enough, though I don't think it's due to parsing responses as my calls aren't even getting through to my dash in order for the API to respond! I just hope they can get this fixed soon. Thankfully I don't rely on these tools too much as I can code, but I moved away from Windsurf, Cursor and Cline due to the unethical way they prey on users for engagement and message count via their system prompts (I have proof and was subsequently refunded, presumably to shut me up). They are literally instructed to cause issues. I'll check in again later, but Deepseek as of now is not working in this fork. EDIT: I have checked my dash and can confirm that the two test calls I made from Roo Code via VSCode were sent to the API (it just took its sweet time). It's possible you're right about response parsing. Doh! EDIT the Second: Well now it appears to be working! This entire thread is a result of a patchy API failure! Which frankly makes me laugh. Good news at least! |
The Deepseek still does not work for me :( |
|
Me too; I guess DeepSeek is experiencing technical problems. |
|
Sorry to hear that chaps. All services appear to be operational on the status page https://status.deepseek.com/. I'm in UK and it's working here currently. Does the status page show anything different to you where you are? All I know for sure is that this is definitely not a Roo Code issue. |
|
Mine is back since a minute ago. |
|
An error occurs again |
|
same problem |
|
Just thought I'd check in again! I now cannot access it! It's 100% Deepseeks end. We all just need to wait I think.
As of right now their API site is down. Some upgrade in progress. |
|
feels like a war ;D |
|
Curl is working for me now, but both Cline and Roo Code still aren’t. I’m using both the V3 and R1 models directly from DeepSeek, and it seems like they might be blocking requests or something has changed on their end. |
|
100% malicious attack. Makes me more inclined to exclusively use non Western models. Disgusting behaviour. |
Same issue here. (Used to work 2 days ago). (@JB-SelfCompany . By the way, How did you access logs ?) |
no, i dont have access |
|
Deepseek has now started to change little bits of code here and there and is now destructive. This NEVER happened even once before this obvious Hegelian Dialectic started to occur. Looks like the West didn't like that and have forced them to increase message count and engagement by introducing the same destructive system prompts present in Claude, GPT et al. It was good while it lasted. I really need to sort out a proper beefy local AI server. It's the only way you'll ever get them to do what YOU want. What a shame. |
Which version of the app are you using?
v3.3.2
Which API Provider are you using?
OpenAI Compatible
Which Model are you using?
Deepseek-{V3,R1}
What happened?
I am constantly getting the error
API Request Failedduring conversations, the error reported isCannot read properties of undefined (reading '0').I am not sure but I think it might be because of parsing the streaming response?
I am able to get multiple responses successfully, but it gets to a point in the conversation where all requests start to return the above error and I need to start a new conversation.
Steps to reproduce
Relevant API REQUEST output
Additional context
No response
The text was updated successfully, but these errors were encountered: