Bolt.diy vs bolt.new and configure codestral v2

hello, i am experiencing problems with bolt uploaded on cloudflare, i try to upload components taken from 21st.dev, but in no way i succeed, it gives me problems with vite, if i try to copy the same prompt taken from 21st.dev on bolt.new instead, it goes without problems, this is making me very frustrated, how come there is this abysmal difference between the two bolts ? also how come when i use bolt.diy uploaded on cloudflare it consumes me big compiutational resources that of energy of my computer ?

One last question, how do I configure codestral v2 with bolt.diy ? I have already put in the cloudflare project variables mistral and its API key but nothing

thank for your supporting

Hi Lupen

Bolt.new is fully optimized and becoming quite mature for it’s still young age. Bolt.diy was only forked around mid-October 2024. It’s been designed to provide more freedom in configuration but it still needs more time to develop it. There are some great enhancements about to be rolled-in to the project in the coming days and weeks. It’s very important you understand many of the Provider / Model pairings have limitations. Some are free and some cost. There are new LLM’s that are better optimized for coding and their size and context lengths are improving quickly. My preferrence currently and for many others is Google Gemini Flash 2.0. And I run in Cloudflare with no issues. I can also run locally on localhost installation but seems roughly same results.

Understand there are limitations on context length and how you word your prompt. If you’re also importing a project. Once you get around 70-85 files in size, there is issues with prompt completion. This is another enhancement coming soon.

I’m not sure I understand your energy usage on your computer. Do you mean CPU usage…?
That shouldn’t be an issue. Unless you’re running a local copy and trying to run Ollama or something which requires heaps of DRAM.

With your last question. What did you mean by “cloudflare project variables”…?

Do you mean the Provider and LLM as below…?

image

To get an idea also on Providers and LLM’s performance and their edit format (diffs or whole file - this is important when looking at Token usage and context size)

Aider LLM Leaderboards

See also: Bolt.diy vs Bolt.new

1 Like

I use an Intel MacBook pro and going into the activity monitor I see that it uses a lot of CPU and starts the fan overheating excessively, also the battery dies drastically, we are talking about an hour or so and I lose all the battery…
I only use the version uploaded to cloudflare
By variables I mean the API key configuration within the project settings you cloudflare, I entered MISTRAL_API_KEY and entered the chaive of codestral v2 but it doesn’t work…
Anyway I currently use gemini flash 2.0 or sometimes deepseek although honestly it is not convincing me, I prefer gemini by far.
Being a designer and coming from webflow, maybe at the moment I am better off with bolt.new ?
I realize that it makes me very frustrated that it is impossible to get what I want and how I want it, I would like something that is very flexible and above all that facilitates me without going crazy.
Between bolt.new v0 or lovable which one is better ? I hate message limits… hate limits in general, no limits in life !!! :man_supervillain:

Watch @ColeMedin s newest youtube video: https://www.youtube.com/watch?v=yRIEpNlacd0

He showing exactly what you ask.

1 Like

Great job, I have been following your videos for a long time but, many of my questions still remain unanswered

Thanks, but as said in the videos. Give me feedback whats missing, what topics to cover in next videos and then I can think about how to do. so far no feedback from users, that stuff is missing or other videos needed :wink:

1 Like

Codestral V2 doesn’t work under the Mistral api. It has it’s own API key. And it doesn’t work with the Bolt.diy api section yet. Here is a video of how somebody got it working in Bolt.diy around the 7:11 mark.

@sfason65 How you come to that conclusion?

Codestral is available in Mistral Provider. This represents always the latest available Model from Mistra, which is Codestral 2501 (alias v2).

You can also see here: Models Overview | Mistral AI Large Language Models

Its just not needed what he tells on the video :wink:

Sorry if I am in error. I made an assumption partly because of the video and I couldn’t get it to work. Also in the console.mistral.com they have a separate section for Codestral that states “Use Codestral via your favorite Code completion tool for free.” that has it’s own API separate from the API Keys section.

@leex279 Were you able to get Codestral free working under the Mistral API section in Bolt.diy?

Lupen

I assume you mean in the Cloudflare setup

You entered the API keys in as per below step

You can always modify the key manually at the prompt if you’re not sure you entered it incorrectly.

What did you mean by ‘chaive’…?

I can’t see Codestral V2 listed either - and I’m on the main branch.

Please try and be specific when explaining any issues or requests. It makes responses easier and faster. :slight_smile:

I found the following which clears it up somewhat. Mistral uses a different domain endpoint and API to use the currently free version of Codestral V2.

@sfason65 thanks, but I´m still not sure about this. Cause api.mistral.ai has the same model.

@thecodacus maybe you can help out?

to change the api endpoint we need to either add it as separate provider or make the base url configurable.
or @sfason65, if you adventurous you can change the baseUrl yourself

@thecodacus my question was more related to if there is really a need to change it or if both apis provide the same model, as I cant really get this from their docs.
As I read it, they do provide the same model, just have different endpoints depending where you want to use it and also different rate limits/payment stuff.

This is what AI gave me:

i believe they have this for codestral models and has free credits. I am not sure if there is any pther changes apart from pricing

1 Like

@leex279 @thecodacus Thank you for taking the time to respond. Since it seems the free use is limited to 8 weeks I assume from their 1/13/25 release date, it probably is not worth the effort to integrate. I found I could access Codestral 2501 on OpenRouter. I ran a quick test to see if I could notice a difference between the two.

I used the The Fiscal Code challenge at https://edabit.com. It is under the expert level JavaScript challenges. I then started a new chat using the standard Mistral Codestral selection. The first iteration of the code came back with 1 pass and 4 fails. I pasted the results back into the chat for two more iterations with the same fails. Further attempts broke the code completely.

I then opened a new chat and selected OpenRouter Codestral 2501 and did the same process. The first iteration came back with 3 passes and 2 fails. After pasting the results, the next iteration yielded 4 passes and 1 fail. Further iterations did not improve and ended up breaking the code again.

Interestingly, I reverted the chat to the code with 2 fails and switched to Deepseek coder. I pasted the errors in the chat and got 5 passes on the first try.

I know there are too many variables and unknowns to put much stock in my test. I just thought some might find it interesting.

2 Likes

ok with codestral but what about overheating and too much battery consumption ? not possible ? i am using Arc and Bolt.diy on cloudflare, where am i going wrong ?