bolt.diy: Deepseek not working on existing projects
Describe the bug
Thought initially was a problem with the api key, however after running some tests, I can see the issue coming from the existing project I’m working on. So if I’m to start a new project from scratch this I would not encounter this issue.
A bit of background: Started the project initially on Bolt.new, moved to the local version of Bolt.diy, continue the work using Google Gemini API 2.0 - relatively fine, then today I tried to use the newest version of Deepseek v.3 both coder/chat and it doesn’t work, getting this error when using the prompt:
``
Link to the Bolt URL that caused the error
localhost app
Steps to reproduce
- Load a current project you are working on via Bolt.diy
- Change the LLM to DeepSeek API
- Type any request, or simple ‘test’
- Error shown via the toast message (see screenshot)
Expected behavior
to work?
Screen Recording / Screenshot
Platform
- OS: [macOS]
- Browser: [Chrome]
- Version: [latest]
Provider Used
No response
Model Used
Deepseek v3 Chat / Coder direct api + via OpenRouter
Additional context
No response
About this issue
- Original URL
- State: closed
- Created 2 years ago
- Reactions: 8
- Comments: 34
This is most likely a problem with the API key you using or the provider itself. For Deepseek e.g. you have to pay credits to make the API work (its not free!). For OpenAI you also need to pay at least 5 Euro I guess to get to Tier1 and usable API.
The main branch has extended logging for the terminal (new features). Check this out, try your prompt again and check the terminal for errors.
Look it works, if you were to start a new prompt chat from scratch, and is quite fast compared to Google Gemini but no matter how you try to use it on an existing project or start new project and import a folder then it fails to do anything.
Reply @mbenhard
Actually its configuration seems right as Deepseek DOC says output context length is 8k not 4k also the bolt.diy configuration correctly configured to 8K
Not sure is because of that
No. I created my project on bolt.new. I downloaded boly.diy, imported my project inside, configured my API Key with Deepseek (i have paid the API Key) and there is big problem. It doesn’t write entirely the content file, it doesn’t modify always directly the files but write the code in the prompt, there is often errors on the code returned…
I get this Issue when I tried to run on a existing project, the request gets around ~140k/tokens, as paliative solution I deleted package-lock.json and get around 11k tokens/request.
Seems like the solution is to use cline or something else since this unfortunately isn’t being addressed 😦 wanted to like this project!
Same issue here. I think it might be due to the low input context window (64K when using their API). I might be wrong though.
Also noticed that by the default DeepSeek sets maximum output length to 4K and max_tokens value needs to be set, not sure how it is configured in bolt.diy
I dont think it works too I tried this method but error remains
yes facing the same issue. Thanks for bringing it up.