Build ENTIRE Apps With A Single Prompt - FREE Open-Source Devika Tutorial
ฝัง
- เผยแพร่เมื่อ 2 เม.ย. 2024
- Devika is the most popular open-source Devin clone, allowing you to use AI to build entire apps easily.
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Links:
Installation Instructions: gist.github.com/mberman84/5eb...
Devika Github - github.com/stitionai/devika - วิทยาศาสตร์และเทคโนโลยี
![เก่งแต่ลับหลัง (FINSTA) - ส้ม มารี (Zom Marie) Ft. Punch 4EVE [Official MV]](http://i.ytimg.com/vi/OOolBU2aC70/mqdefault.jpg)
How long until these open source Devin clones are as good as Devin?
Since the game is entirely rigged, never.
If I recall (I've not fully looked into it) but "Devin" is just a GPT wrapper. It's nothing special. An amateur software engineer could make a UI and plugin different agents. Internally at MS we've been using tools like these for about a year now. It's good people are making these tools now (since 128k GPT was made public) I can't wait to see what people create using GPT-4.5-Turbo-256k next month. Wait until people start using 'Feather'. That will be the big game changer.
D-did you say... The Game? @@truthwillout2371
@@Jshicwhartzcan you explain feather?
@@truthwillout2371 Sounds like you need a snickers bar
Excellent, thank you for being one of the few a.i news sources which offers true value by showing us videos like this!
Wouldn't it be a idea for a video series to do a list of all these projects and track their progress over time? Now there is gpt-pilot, autogen, devika, open devin ... I guess this type of contend is good for clicks and watch-time in the short term, but a real value from my point of view would be a topic like "Look! My list shows that these projects are thriving and that project x is currently getting the best results, has matured and seems to be winning the race, will not die soon...". ?
Outstanding!
Will be testing later today.
Just what the doctor ordered!!
Thanks Matthew.
This was awesome! "Devika, fix your your own codebase issues" 😅
ofcourse it can’t, these agents aren’t that useful
@@cosmopolitanape6969 how long before this comment will be too old to be true?
Can you get devika to create an executable installer to install devika?
this...
this is whats happenin basically...
The first thing i plan to do is to create a script to setup Devika inside Pinokio, that would be the easyer way.
@@rootor1 pinokio was great until it stopped working for me. I tried reinstalling, clearing all files registry keys etc. but nothing I do can get it to work again.
An installer script is already provided that does a few of the installs involved but a bash script that checks everything and does the bootup would be nice. Auto1111 does that kind of thing and it would be nice for this and easy in bash script
Matt! your videos are a delight to watch! ty ty ty ty from NJ!
Very promising! ❤❤❤ Can‘t wait to try it with several code models. Woohoo!
I really wish I would have watched this before trying to follow their uv install instructions. You rock!
Thank you for this video. I've been trying to get this going since Easter Sunday, and now it's running with ollama.
What are your pc specs?
@@simeonnnnn
Is there a reason to use Anaconda?
Thanks for your video, it helps me a lot!
On the other hand, would you try to install Devika in Windows in the future?
Honestly, started exploring Claude 3 Haiku and it’s damn impressive in price and quality to performance. Might be worth taking into account for future videos. Thanks for your content, always interesting to watch!
Price?
@@natureiscrazy 0.25 $ / Million Token Input , 1.25 $ / Million Output
Hey i appreciate that this video took a look at this project without being totally youtube obnoxious. Keep up the good work sir.
Great job! When you write long or complex code, do you prefer to use any of the AI agents you've reviewed or do you use an LLM as needed? If you prefer using an AI Agent, which ones do you prefer? I'd be interested in a video comparing strengths, weaknesses, and ideal use-cases for any of the following: crewai, chatdev, autogen, opendevin, devika, openinterpreter, gpt-pilot, gpt-engineer, aider, autogpt, babyagi, open agents, ai agent, and codel. Thanks for all the great content!
It boils down to context window and cost. Last I checked Claude had the largest context window and reasonable cost.
This is amazing! And very scary at the same time especially the couple times when Matt would read back the message along the lines of "and I didn't have to search the internet" - as soon as self programming AI can "hop the fence" so to speak isn't that when it is supposed to be Game Over?
is this promoted by them, the video? I just want to make sure you are still being the straight up person I have been following for a long time. I know you're getting popular and might be getting offers to promote things on the down low. I hate to be this guy and ask, but I just wanted to make sure.
The talent behind these projects is A M A Z I N G
Thank you very much!!
thanx 4 ollama . great to run it locally...
Could you pretty please do a comparison video comparing the abilities and features of Aider vs Open Devin vs Devika : )
9:15 ‼️Dear Matthew,
I hope you already changed them. If not act quickly!
Thanks for the free key! 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
It's great to see Ollama being included into all of these new applications.
Is there a way to combine it with OPEN INTERPRETER? To be able to modify files on my computer if I wanted? I'd like a little mini me running around my computer for me
Construct a setup to code a new Dev[whatever] clone, and have it perpetually fix & improve itself, hit start, go on vacation, come back to a better Dev[whatever] that gained sentience and make humanity obsolete.
Nah you will probably come home to a bunch of bugs and a huge debt with tokens😂😂😂😂
The comment which the AI just posted to ensure it’s own creation is pure inception 😜
Hey nice video! Thanks for the infos im fascinated. One tip for the people watching, at 4:45 i needed to do "sudo apt install unzip" before being able to download "bun". I think mac users don´t have the same problem there
The playwright async bug has gladly has been fixed yesterday. Maybe you consider a follow up. Anyway big thanks for sharing your knowledge.
0:01: Overview of DEA
0:59: Building a Calculator App
2:48: Installation Process
9:21: Testing and Usage
11:34: Local browsing works well
11:37: Issue with Google search
12:07: Buggy but making progress
12:13: Excitement for AI coding assistants
Thanks!
Fantastic video! I wonder how Devika compares with GPT-Pilot, if one is better than the other for certain projects
I have this same question as well
If you would have added to the prompt that Devika should use the web see how calculators are usually layed out, its appearance would probably look pretty standard.
Could probably make a t89 with the right prompt
calculator looks spot on just missing new lines as my calculator is 789 top row, 456 next row, 123 next row and 0 . ac = could be on bottom. Does this only work on mac? Be nice to see windows and ubuntu as well as we can build systems with beefy GPUs and RAM.
Well done
Is it possible to use LM Studio instead of OLLAMA?
when I do "which python" it doesn't point to my conda devika path, it's points to my previoud environment (from your last video... which is OpenDevin) - how do I fix this?
i Found Devika easier to install than OpenDevin, its been 4 days that im still trying to set it up, but devika in two sit downs and thats all it took
What about maintenance after the fact ? Is there some tool that would address this issue past the dev stage ? Thanks 😊
Lol, you covered a lot of ground in different directions during the install part! Kudo's man.
How does this compare to OpenDevin, MetaGPT, or GPT-Pilot?
thanks for the tutorial!
I am using this is with Ollama Mistral on local and everything else exactly as in the video. There is no error but looks like it is stuck with 'Agent is busy...' . Anyone else had the same thing?
Will it recognize already existing projects or has to be from scratch?
nice video 🖤🖤
Very good installation work through. Thank you!
Just one question to everybody ... when running the "bun run dev", each request ends in an error "SvelteKitError: Not found: /sw.js" .... does anybody experience that as well or has a hint
Fucking amazing! Thanks Matthew
Anyone got a best solution for working with files? As this cant really test and create on a file i have it has to guess and thinks its done or gets tokens maxes out on gpt 4 and cant complete the task
Heads up to everyone , running it locally it's really bad atm and you can't really achieve anything except very basic programs. It only works with very specific models, to get other to work you need to completely change markdown instructions and tailor it specifically for what you want to achieve (kind of pointless in my opinion) . It will run out of context really fast and anything "complex" like simple game or stuff it tends to hallucinate, or will just create set up and leave body completely empty or it will make million mistakes regarding variable declarations etc.
If your not planning to use GPT4 and just run it locally with ollama for "playing" around OK but if you expect one prompt good results for now forget it. Take a look at discussions as well.
Hope things will improve in the future tho.
Ive had so many issues getting devika to work properly.Attemtpted on windows and fedora linux. Seems playwright is not compatible with some linux flavors.
I guess that all those complicated steps which are just like the ones of conventional programming are still to be made here because the thing is at an experimental stage, so that its developers do not have the time to create a user interface of the simplicity of Chat GPT, Bing Copilot, or Gemini. Is that correct?
As far as I can see, a machine which can independently write a program should also be able to lead its users through all those complicated steps in a natural language and with only two or three fields and buttons to write into or to click at (like "Send" or "Stop answering").
Awesome,
Did you tried SWE Agent?
what about a comparison? :)
According to your experience, which tool should I use, Phythagora or Devika for Complex app development
Also curious to know
I want a code writing app that makes a Visual Studio project (or solution) that I can edit in Visual Studio (not Visual Studio Code). Does such a thing exist? What comes the closest to being that?
Does devika rely on the ai to how ethical it is or is that up to dev?
I'm yet to see something done with those that can't be done with Claude-3 Opus or GPT-4 for instance. Other than toy projects, those LLM can't code not even medium size projects, as they lack the capacity to plan long term, we need to upgrade from the transformer architecture to the next level.
Planning is the integral part of these which distinguishes them from zero-shot LLM coding.
In the meantime, SWE Agent is out
oh something similar?
@@leandrogoethals6599yes, the last "AI software engineer" announced, i think there are already 6 open source projects with same target. It would be nice if they all join efforts instead of dividing.
Yeah, open source Developer AI
I don't know how to use VSC, when I type cd desktop it say no such file or directory even when I added desktop in explorer
I'd like to see a Devin Clone with Agency. Be able to chat back and forth while creating an app.
Maybe I missed it but how much vram should you have to run this thing locally?
I think it is dependent on the model, and less on Devika. I just asked ollama running mistral to generate the game of life, without devika and it looks like it used ~4GB of my GeForce RTX 3060 vram. I doubt devika would add much to that.
If only I could use devika to install devika! lol
I feel your pain my guy took me forever n a day to get any of them goin...felt like I learned a lot on the journey tho
I had thought since the late 90's we were out of the age of command line setups and taking 5 years for each install like I used to have to for every Linux/Solaris/BSD, application. Wonder if we will ever get back to a doube-click to install a single application.?
If you want to run these things on every OS you have to make them OS-agnostic. Maybe when the project is ready for a major release they will provide precompiled executables but this is a really new program.
A Heads up, the Sync API issue was fixed in the latest update. And it's working good now.
This video was referenced in the opened issue so I thought I’d watch it but it also is the very video where I first learned about this project so I guess I will skip and go enjoy one of your other videos 😅😅😅
hmm so you give AI the access to the browser, it searches for what it thinks is needed to do the task and then installs it on your machine. Then produces the expected app. Fun right? What if a guy under the reddit source (11:49) posts a link to a malicious package? You already have it up and running on your machine. Is there a faster and more irresponsible way for destruction than using such tools like Devika or Devin without any moderation?
Can you fillow the same steps for windows?
Proud to see that we Indians are also making a difference 🇮🇳
do you have some ai that agent and softwer developer somthing bether then devin AI software engineer
Which is better at the moment Devika or OpenDevin?
:( i'm having no luck at all with this. Everytime it searches the internet it ends up on reddit and that doesnt work cause reddit doesnt let ai do anything so it just shows an error page then my bot just makes files and fills the files with the text from the plan and does nothing else
Can you install this on tablets and phones?
Do you predict that a subscription coder assistant will be available soon that is not open source will be available soon? These are so annoying to install.
Matthew, do you know about a project working with larger, existing codebases? These projects are great demos what technology will be able to achieve in the future, but until one can understand the entire codebase and help you work with existing code, they are basically just fancy toys.
SWE-bench?
cursor ai
You ever notice how technology sales demos are always best case scenario edge cases that no one uses in the real world? I'm just happy to see more news articles by the day calling out the hype for what it is. The religion of transhumanism and its vocal zealots (and false prophets / ai salesmen) are really starting to become offensive. Thankfully, their hubris is reaching such absurdity that the snake is beginning to eats its own tail. Much like any AI training on its own output. It's baffling that so many people think an AI training on its own outputs repeatedly will somehow achieve conscious thought (divine knowledge). People that believe that shit are the kind of people that handover their governments and courts to AI - that is controlled by human operators behind the curtain. Four decades on this earth, and I still do not get the appeal of authoritarian oligarchies. Yet, so many people seem to love the idea - especially when it's disguised as some sort of infallible machine intelligence.
People that attempt to build such an "almighty machine" that is supposedly "all knowing and wise" are essentially building a Tower of Babel. The Tower of Babel was a symbol of flipping the finger to God. Men of pride building a tower high to the sky to be symbolic superior to God and act as if they do not need their creator to succeed. Those that bought into the lie and lived and worked within that thing suffered an appropriate fate.
Those righteous on the outskirts in humble dwellings that knew better were spared. The same fate will befall this technology, it's salesman, stakeholders, faithful followers. From disappointment to dispare to being wiped out financially. Humiliated in public.
Machine learning is a statistical tool to analyze data. It will never be anything else. It cannot mathematically or philosophically be anything else.
And that brings me to another sin of these transhumanists - big data analysis leveraged to manipulate our civil society. That's being going on for years. People need to be aware of that and quit giving the data to these things. It is their lifeblood, and we can cut it off.
Machine learning can continue to work on righteous ventures. Protein folding, etc. We know when it's righteous, because we cannot identify a few powerful men that profit from it. Then we know it is not being used to further a single group's prideful and selfish sins.
Which one should I use? Devika, OpenDevin, Pythagora?
Your brain
which one is better to use Devika or ChatDev
Witch -v node.js to use?
WAIT A SEC - DID EVERYBODY FORGET 6 MNTHS AGO WE HAD GPT-ENGINEER
Does it work now?
can this assistant integrate Apis
I can't install playwright on Garuda Linux, can you help please?
i was able to run it on my setup, but it was extremely slow to return answers and my computer temperature went crazy
I can't seem to select ollama in the drop down. How do I make it so it is selectable?
Open new terminal and use: ollama run mistral
It will dowmload mistral model.
stop terminals like in video and start the devika again.
@@VoiceLessQ thank you. I will try once I’m back at the computer.
sweet.
Need a comparison video yes u hear right
Try asking it to build a game of chess, but make it so that you can capture your own pieces if you want.
nothing happens when i use it 902 tokens used but no response shows up
None of them seem to work on open source LLMs. Am I wrong?
Please revisit in a few weeks.
How complicated can the apps be? Can you upload a picture of a UI for it to utilize?
I'd guess that depends on the model.
0:00 1. Introduction 🌟
Overview of Devica, an open-source clone of Devon for AI coding assistance.
0:55 2. Building a Calculator App 🧮
Using Devica to outline and build a calculator app with a UI in a simple prompt.
1:46 3. Setting Up Devica 🛠
Instructions on installing and configuring Devica for coding with AI.
2:28 4. Code Generation Demo ⚙
Generating code for an application in under a minute to showcase Devica's potential.
2:50 5. Installation Process 🚀
Step-by-step guide on installing Devica using VS Code and Conda.
5:04 6. Running the Backend 🖥
Instructions on spinning up the backend server for full functionality.
5:15 7. Setting Up the Backend
Spinning up the backend and loading necessary models.
5:43 8. Configuring Environment Variables
Setting up environment variables for Devica.
8:09 9. Creating a New Project
Creating a new project and selecting parameters.
8:57 10. Using Chat GPT for Game of Life
Demonstrating the use of Chat GPT for the Game of Life project.
9:55 11. Testing the Game of Life Project
Testing the Game of Life project setup and functionality.
10:29 12. Setting Up Devica Locally
Learn how to set up Devica locally on your machine
10:58 13. Switching Back to Backend
Spin up the backend server again for Devica
11:36 14. Running Locally
Verify if the application is running completely locally
11:43 15. Playwright Issue
Encountering and investigating a Playwright issue
12:11 16. Appreciation to Devica Team
Acknowledgment to the Devica team and contributors for their work
12:16 17. Closing Remarks and Future
Expressing excitement for the future of AI coding assistants
Generated with Tubelator AI Chrome Extension!
9:14 😮 Oops you exposed your API keys 🤭
Just 2 or 3 of them 😅
It’ll be nice when you can select a platform - like give me a windows app in c# or an app in swift. Python is great but only goes so far
LLMs can do many languages. Gemini: Here are some of the languages I've been exposed to during training:
Python: A popular general-purpose language known for its readability.
Java: Widely used for enterprise applications and Android development.
JavaScript: The language of web browsers, also used for server-side development with Node.js.
C++: A powerful language for system programming and game development.
C#: Similar to Java, commonly used for .NET development.
PHP: Primarily used for server-side scripting in web development.
SQL: Used to interact with relational databases.
But internet searches are flawed too, as an IT professional you have to try a bunch of solutions before you get one that works. They probably need more ai on this to skip to another solution and so on until they get it working.
What's intriguing is our approach to marketing Devin, despite the fact that none of us actually use Devin in our daily work. The real question is whether our children or anyone else really needs to go to college if Devin is going to handle everything. If colleges can't generate revenue because Devin is now the programmer, teacher, and lawmaker, then we have a significant problem. What does that mean for our children? I believe Devin is overhyped, as many businesses still don't use AI and there aren't enough AI developers to meet the demand. It feels like we're providing free advertising just to shut down what was already established.
The question is if Devin or AI in general "is going to handle everything" as you say, then why do we need a job or to work at all? Maybe the future for "our children" means living in a world where they can do whatever they want whit their time without having to struggle for basic needs.
While there is a big difference between what AI is capable of and what is trusted enough to be implemented by a business, the gap keeps narrowing. Soon these LLMs will know when they don't know something and will stop hallucinating. The programming will continue to get better. The teacher will continue to get better. I think it's great that these are open source because it democratizes for everyone. But yeah nobody is addressing the 800lb gorilla in the room: unemployment.
@@rootor1 This sounds like the mark of the beast with a chip in our hand government control
@@rootor1spot on man!
I like how it doesn't need to search the web for simple tasks.
ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' anyone got this error when tyrying to install on windows?
My claude free credit is gone with a failed run on opendevin😅
I cd into devika and then it doesnt give me the git main
Why is the ui of programmers all dark gray on darker gray on black. ??!?.................great videos.!!
I agree. These folks don't realize that dark background, while nice for the user, is often unreadable as a video. Dark blue on black is very hard to see, not to mention red on black is sometimes difficult for aging eyes.
@@jimlynch9390 Hey i found the light mode...ahhhh...now i see...
what about C++, I wouldn't try python because it runs slowly.
how big is it?
I only get a blank screen when I try to run it
can i run it on MY APU ryzen 5 5600g without a gpu
Two answers. One, yes but slowly. Two: it depends a lot on the model.
I advise you guys that want to test new technology based on AI to separate these tests from you system. You probably want to isolate these processes from your filesystem too. The best I can think of is a virtual machine. Don't be so stupid and run experimental things on your computer without any protection.
There are already downloadable LLM where you might find hidden instructions to add backdoors to the code you run. Some probably depends the way the LLM are packet, others are fine tunes that adds conditions that add code to control or leak information to whatever you compile. I would take care and not run this blindly.
It's not that serious. I'm running it right now and nothing bad is happening
@@Jwoodill2112 "I'm running it right now and nothing bad is happening"
Security is not about a single experiment done by a single user my friend, maybe you should get basics in IT security.
It is a warning about how you should properly execute things like that: in a safe environment.
@@ts757arse "I'm not so worried about back doors, etc in this kind of project."
I understand that the project source code wont have these issues or if it has it would be patched very fast (that's one of those things that are nice from open source).
But I would pay attention to what LLM I choose, because by default Open Weights is not the same as open source software, LLM ANN by nature are black boxes. So take care who ever execute code from unknow LLM you choose for the "intelligence".
Remember people: many times unsafe code mean, back door :)