Claude-3.5 Sonnet + Qwen2 + ContinueDev : This AMAZING Coding Copilot BEATS Github Copilot & Others!

In this video, I'll be telling you that how you can create a new coding copilot / assistant with ContinueDev, ShellGPT & the New Anthropic Claude-3.5 Sonnet Model. This new Anthropic Claude-3.5 Sonnet model based copilot is cheaper and better than Github's Copilot & every other local, opensource copilot I have ever seen. This Github Copilot alternative can work even without the need of GPUs which is really amazing.
This alternative can also be used with any opensource LLM, OpenAI models or the other Claude models such as GPT-4O, Claude-3, CodeQwen, Mixtral 8x22b, Mixtral 8x7b, GPT-4, Grok-1.5 & Gemini Code Assist.
------
Key Takeaways:
🚀 Unleash the Power of Claude 3.5 Sonnet: Claude 3.5 Sonnet outperforms GPT-4O, Gemini, and Llama 3, making it the ultimate copilot for coding and more.
💡 Integrate with VSCode Easily: Using the ContinueDev VSCode extension, you can effortlessly integrate the Claude API for enhanced coding productivity.
🔧 Save on Costs with API Efficiency: Claude's API charges are based on usage, making it a cost-effective alternative to GitHub Copilot's $10 subscription fee.
🌟 Dual Setup for Optimal Performance: Combine Claude 3.5 Sonnet with a local model like Qwen2 1.5b for chat and autocompletion, optimizing speed and reducing expenses.
🛠️ Set Up ShellGPT for Shell Commands: ShellGPT provides a seamless interface similar to GitHub Copilot's Shell suggestion tool, enhancing your workflow.
💬 Enhanced Coding Experience: The Claude model can generate code, insert it directly into your files, and even offer context-aware suggestions, boosting your coding efficiency.
🎉 Community-Driven Development: Join a community of developers switching to Claude, leveraging its superior capabilities and cost-effectiveness.
-----
Timestamps:
00:00 - Introduction
00:11 - About Claude-3.5 Sonnet
01:00 - Why Github Copilot is not worth it
01:18 - Making a Custom Copilot with the Claude-3.5 Sonnet Model
01:57 - Using Local Model for Autocompletion
02:18 - ShellGPT
02:35 - Installation (Starting)
02:54 - Installation (ShellGPT)
04:02 - Installation (ContinueDev + Claude-3.5 Sonnet)
06:17 - Installation (ContinueDev + Claude-3.5 Sonnet + Qwen2)
07:33 - Conclusion

[@aricoleman5802]

Great video. Would be awesome to see more like this and other models used if possible

[@ryanrrothwell]

Thanks for making these videos.

[@mjherich]

Excellent walkthrough of how to set this up 🙌

[@TheTimtimtimtam]

Thank you for this video, much appreciated

[@dmitriipukhov3887]

love this! thanks for sharing!

[@garthh7314]

this worked! you're a legend!

[@LiverTGP]

That's an amazing content! Which LLM do you consider to be the best for local use in the ContinueDev chat? Do you have a video on this? Thanks :)

[@2STR4NGE33]

Thank you for this!

[@NormTurtle]

this was the most easy and hands on explanation.

[@tzardelasuerte]

How about using Claude 3.5 sonnet with gpt pilot Pythagora?

[@agush22]

cool, what OS are you running?

[@Miamisinhijos]

Excellent: 10/10

[@frustasistumbleguys4900]

why i always got missing config key for openai?

[@geepytee]

Interesting, but why go through all of these steps instead of using something like double bot, which just works out of the box and has all of the right models?

[@DinoRossYT]

Is possible to handle questions to the web interface from vsc? instead of api i think the $20 plan would be cheaper for our use in much scenarios

[@SahilP2648]

What do you mean? Using the extension if you just use the pay as you go model, then yes, it's not for code only.

[@tobeyforsman]

Pretty cool

[@vitalis]

So cool if I knew how to code :) is there any setup for Notion?

[@AICodeKing]

Will try to cover something like in upcoming videos.

[@jackflash6377]

With Claude 3.5 as the Copilot, can it see your entire codebase? If it can, compare it to just embedding your codebase. The other day I sent my codebase to Claude 3 Opus and worked through some issues I was having. The API costs were $64. I'm wondering if the codebase got embedded into the prompt and was sent back and forth with each query. How else would it use up so many tokens?

[@AICodeKing]

There's option to add your entire codebase as context in ContinueDev. But, I wouldn't recommend that due to the huge API costs that might come. Instead, you can just add the context you need to work on. Also, The new Claude 3.5 Sonnet is much cheaper than the previous opus model (Previous one was like $15/m token while this is just $3/m token)

[@SahilP2648]

@@AICodeKing btw github copilot uses GPT-4 or so they say. Sonnet-3.5 is really good but is it going to be worth it using like that (as in pay as you go model)? Also, github copliot is privacy oriented and secure (or so they say) compared to Sonnet-3.5 which is not private (they will work on the data you provide afaik).

[@2099EK]

@@SahilP2648 They use GPT-4 for chat, but GPT-3.5 for everything else.

[@enkor349]

@@AICodeKing could I add files and knowledbase (entire codebase and related similar projects) to an assistant in OpenAi playground and then use that particular assistant. Would that bring down costs since does not need to send codebase back and forth?

[@hineko_]

AiCodeKing sounds like the goth guy from the ITCrowd :))

[@tejasness]

Groq API + Continue ❤

[@ColinNardo-le3bl]

any reason why sgpt doesnt seem to work as a command for me

[@AICodeKing]

Check If pip package path is in your Terminal environment.

[@1TanishChavan]

Im not able to find the shell_gpt file you showed can you tell the location. Im using anaconda

[@AICodeKing]

It should be at ~/.config/shell_gpt/.sgptrc

[@Papibalada]

It was working great for me but now after a few hours Im getting a HTTP 429 too many requests error

[@cmosguy1]

i had the same issue

[@fabrizio.dipietro]

Hi, how much are you going to spend like that per month? Less than $10?

[@AICodeKing]

It costs about $3 for a million tokens. So, You could get it below that amount if you don't use it for autocompletion.

[@apester2]

@@AICodeKing the problem is that I don't know how many tokens day to day coding will use. Maybe more, maybe less. Maybe even using it for autocomplete works out cheaper, or not much more expensive. Please make a video in a while to let us know what the costs turn out to be.

[@d.d.z.]

Surely less than $10. I've do the comparison last month with Chat gpt

[@apester2]

@@d.d.z. lets hope. I would love to just use sonnet 3.5 for aurocomplete as well. Almost never use the chat and other functionality.

[@d.d.z.]

@@apester2 try it and come back to talk about

[@AXYZE]

What would you say is better, Continue or Cursor?

[@AICodeKing]

I prefer contiue because it's free

[@ernstmayer3868]

Sadly, autocomplete does not seem to work. I installed the model and successfully tested it from console. I adjusted the config.json file to contain the tabAutocompleteModel property with all shown values (it was not there after plugin installation). I see no errors, continue tells me: "index up to date" in the status row. But nothing happens in my code files.

[@ernstmayer3868]

I even tried it with ollama server to spin up the API (ollama ps is showing the model as loaded). No success and no http request in the developer tools of vsc.

[@AICodeKing]

Try restarting VS Code

[@SilverBackGeek]

It looks like OpenAI has changed the code, I am getting the "BadRequestError: 400 Bad Request" when I try to put in the API Key,

[@CruiiserMakBH]

Yeah I found it won't allow me to paste the API key in this way. Find the .sgptrc file in your C: drive. Navigate to C: > Users > "Whatever your username is" > .config > shell-gpt > .sgptrc Then edit as per his tutorial. Hope this helps!

[@paulyflynn]

I am using Jetbrains AI.

[@FranzAllanSee]

installing a python library globally sounds very windows-y. I know you can do it in *nix. I do it sometimes. Still feels dirty tbh 😅

[@johnrich7879]

You have very nice voice

[@369era33]

need support

[@chieeyeoh6204]

For real? GitHub Copilot still use GPT-3?

[@AICodeKing]

Yes, it does.

[@stianhenriksen4826]

Chat says it uses gpt4 turbo?

[@AnasQiblawi]

Claude in programming is waaay much better than ChatGPT

[@majukanumi9639]

really you are jumping from A to B...

[@uaer309]

Github copilot is driving me nuts. It's sooo dumb.

[@bambanx]

can´t found .sgptrc, thanks for the video

[@bambanx]

found cmd+p ~/.config/shell_gpt/.sgptrc :)

[@AICodeKing]

You can check the ShellGPT github repo. It has the path listed there.