OpenDevin Tutorial (Open-Source Devin) - Build Entire Apps From a Single Prompt
HTML-код
- Опубликовано: 29 мар 2024
- Open Devin is a 100% open-sourced version of the massively popular demo of Devin we saw a few weeks ago. I'll show you how to install it and how to use it.
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:
OpenDevin Github - github.com/OpenDevin/OpenDevin
Install Instructions - gist.github.com/mberman84/2ad...
Disclosure:
I'm an investor in CrewAI 
Наука
Crazy how the foss scene is moving as fast as closed sources I love it
with AI writing the majority of code in the future, there should be a great open source alternative for every app
@@patrickwasp quality line of thought. Code wants to be free indeed....
Race to the bottom and besides you can always rent powerful machines to optimize for lower end machines. Which will make it hard to monetize any model for very long plus we are starting to approach the good enough stage for local models so by next year. Will we have decent capability to do most AI things at home with a easier to use interface.
@@southcoastinventors6583 That interface already exist, is called OpenInterpreter and is also open source, i plan to put all my computer power (4 nvidia cards) to help develop it.
Same! I am absolutely loving the development cadence in the ai community right now!
"Do you remember devin" lol talking about it like its a historical artifact DAMN this space moves too fast! :D And this is awesome news TY
Heh, yeah, every month is like a year of progress. We're on the exponential curve right now and it's crazy.
🤣
@@pvanukoff i would say every week we are living amazing tech times folks 🤩
Don’t even try to use it if you got windows 11, it only does up to windows 10
and when agi comes...ai improves by itself without prompt
Thanks Matthew. I like that you show the app and functionalty before it's installation.
Awesome vid man thanks for this! Would love to see another video of you using OP making a more complicated software/app. I would love to see what this thing can really do!
What a great install tutorial, you even explain what to do if things go wrong 😊
You made a very complex installation process look as simple as possible 🧐. Very detailed and user friendly.
It seems like they changed the entire setup process now. pipenv is really damn annoying instead of just using pip and requirements.
You literally sat here and walked through everything. Thank you.
Brilliant. Yet another I need to try out!
Thanks for this tutorial, glad I havent started with Devika, straight to OpenDevin 😁
To anyone wondering where the requirements.txt file went, they replaced it with a file called "poetry.lock". To use it, run the following commands:
pip install poetry
poetry install
The Poetry configuration is invalid:
- Additional properties are not allowed ('group' was unexpected)
Hey Devin, create a portable self executable of yourself. It should give options to use online or local LLMs
it can. i just tried.
@@revelari5250 hahaha love it
RUclips channel about setting up github projects is getting thousands of views.
Shows me many are interested in using open source AI tools
Very well done Sir👏
damn!! 9 days later and OpenDevin is a completely different thing rn, a video series with updates (weekly +) sounds pretty good, the project dont even has 30 days
Thanks very much one again, it´s a very good video as always, thanks very much for sharing.
This is amazing.Thanks Matthew
Control Z doesn't force quit. it backgrounds the task. If you type jobs you can see it running in the background. Whatever number job it is you can kill. remember to add % to the job number or you can crash your system. So if it's job #1 kill %1 will kill it. Then it's gone. Otherwise you can have trouble grabbing the same port.
lol I had a "WHAT?!?!" moment and was going to comment on the same thing. I dunno why I watch this guy. He's kinda clueless and talks like a valley girl sometimes.
He’s using Ctrl C right, not Z ?!
@@fedorp4713 lmao
Thank you!!!
By the way, you can just type “fg” if you want to re-enter a task you pushed control+Z on. That also confirms it does indeed put it in the background
you da best AI content provider !
Great tutorial sir!!
If you want something that has all baby bugs fixed, has every piece of prompt magic that has been seen in a research paper and agentic workflows then look at AGiXT, not limited to programming, even supports streaming torrented distributed ai models and a lot lot more... total geek project that aimed for the moon and went past the sun, not harder to set up then open devin and a lot more mature
👀
@@matthew_bermanreview it for us plz
@@matthew_berman can this be implemented on linux Arc or debian
If can deploy apps like devin..?😊
@@leilaandthewolves9659 it's a platform that you can use to make any custom "app style" application out of, if you want a devin, you can do that with AGiXT, if you want an influencer or a researcher or an author or an artist... you got all you need in there do to whatever you want.
thank you very useful thanks, I wasn't able to install with docker but I'll try again
Babe wake up! Tutorial on OpenDevin just dropped.
A followup video actually using opendevin with ollama would be pretty awesome. It's not really enough to say "you could theoretically run this locally" IMO.
I could not get this to work on my Macbook Air M2... going to try my PC when I get home. Thanks for the great content!
Hi Matthew, great video as usual. How does this compare to Gpt pilot and AGIXT? Would love to see a comparison video of these types of projects, and also which one of them works best with open source models (and which open source models Excell at these types of tasks running locally)
I have a feeling that this too will be obsolete within a few months, which is absolutely crazy when you think about how long it has existed.
Outstanding 💪 🤖
Control-Z doesn't force stop anything. It puts it into pause state. Run "jobs" command next time you press Control-Z on something
Great Tutorial!
Hey! Awesome video, thanks!! I would love to see:
1. A test of a bit more complex application. Possibly using APIs, where it needs to do it’s own research and copy snippets from the docs. (If this works it would really open the world to it)
2. How to use local open source models (possibly with a test of which model would make the most sense, separated in different vram sizes that people have)
PS: is there a way for example having the coder being a local llm and the reviewer gtp4? Could increase cost efficiency a lot
10:45 Ctrl+Z doesn't "force quit", it backgrounds the process. Understand UNIX more. 😛
This looks like a very promising project
Thank you for the video. I am still installing... started with Visual Studio Code, then git extension, now anaconda. Hope I will be able to use opendevin 😍
It's updated on Github
It is no longer difficult to install as it was in 8:00, it can be installed much more easily by reading the documentation from github.
Thanks for your educational content Mr. Berman
10:35 control-z does NOT quit it, it just makes it go suspend the process to the background. Which means if you run the command again, you started it twice. Which means you are using twice as many resources and the background 1 can block things the newly started 1 needs !
Great video. On my Arch Linux with Ollama it worked, without problems, but LLM_API_KEY *must* be set even if Ollama doesn't use it, otherwise the backend hangs.
Yes open Devin!
I love how the world has changed where the force to share
thanks for your content.
Great video! Suggestion: for a real use case - how do you work with an existing repository? Is there a tool out there Devin-like or multiagent that would allow for a real-life workflow, where I point to a repository, ask for a specific change (eg. new functionality) be added to the code and the tool analyses it and implements the change? If yes, it would be great to see a video showcasing this.
Excited to know as well if you find anything just reply to this comment
you mention the use of lmstudio could you make a video of actually using lm studio with open-devin
Developers: 'devin is going to take our jobs!'
Also developers: 'Lets make it open source and free!
My god I laughed too hard at this 😂
If they are using docker, I don’t understand why it needs you to install node and requirements.txt. That should be automated and put into a docker/Podman compose file
Absolutely! Isn’t Docker invented to prevent so much trouble when setting up app environments?
yes ...everything should be inside container .
That project is a total mess right now ...
Yeah that made NO sense @@mirek190 🤣
I’ve started creating a docker compose Pull Request. Looks like the front end uses python to talk to the docker socket, so you have to do some docker-in-docker jank.
Why didn't Devin realise that in the first place? 😂
Ssssshhhh you may kill the hype!
they should put it all in a docker for more people to test.
Thanks your video ❤️
3:35 I feel your frustration. Nowadays days “ programming “ will be based in facing installation problems 😂🙈
my biggest gripe with these types of coding systems is that they only do well for code that already exists. The fundamental flaw is that when you need to develop new code with novel application or algorithms. I guess I’m just waiting for AI to replace my job
Same as Open Interpreter. Can't create anything useful on it's own, but we are getting there.
i got one error at 8:44 saying i dont have the requirements.txt file ! please how can i fix it.
I ran into the same issue, you'll need to install poetry but he doesn't appear to go into that in this guide. I see your comment is from a few weeks ago, any luck?
@@deboe1999 It ran, everything used to look good, but then at the end, in the interface where you ask "build me or help me with..." show me some error, and then I saw all news that OPEN DEVIN was a fake thing, so i let it be, btw i have to uninstall ullama or wherever was the name
@@deboe1999 Same issue when running the command "python -m pip install -r requirements.txt" I get "ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt'" :(
Could you pretty please do a comparison video comparing the abilities and features of Aider vs Open Devin vs Devika : )
AGAIN?! *watches mid shower
Wow this thing EATS UP tokens and $... woah. Even on GPT-4-Turbo, the default, which is pretty cost effective.
I was doing some simple tests on the Calculator concept and project, I'm not too impressed here yet. I'm still wondering/thinking AIDER might be the best/better, but will need to keep looking for now.
I hope this expands to other programming languages, this would be huge.
I'm surprised you don't get a lot of viruses with all these new projects and all their random junk they have you download. But I'm glad you do it first so I don't have to worry about it when I do it
Cool..oh and squiggly line = Tilde 👍
If I have full instructions for features I’m trying to add to a php script for building a social media app will open Devan work or is this another AI tool that only build simple low functioning apps that have limited features?
Great video! Do you have any plans on making a videos about Microsoft Auto dev?
Yes, but is it an actual project yet or just a research paper? I couldn’t find the project
Can't wait the week or 2 before getting a one-click installer of that 😅
Any way to use a Gemini pro API , all of these always require open ai
Matt, it would be great if you show us how to update Open Devin in a week. Do we have to follow all instructions top down, or is there a quicker way to update just the files that have changed?
What terminal do you use? Command prompt, Git bash or PowerShell? I need help; having so many problems as a beginner
This is the first showcase that has made me want to pick up an AI project. So tell me you dont need to have python, docker, etc installed on your PC to use it? Already in the Devin app?
How did you created the docker container? As I can see that you only pulled the image.
great stuff
You should always read your brew messages. You need to link npm. You probably have another installation of npm. That's why it worked without linking.
Ctrl+z suspends a command, doesn't completely kill it. Actually if you type bg then that command will continue running in the background
Do you have any online course or mentoring on the AI Agent Implementation?
When this open Devin can have agents, that seems like it'll be the end all be all
I have installed and used docker before, but is there some prerequisites in this case to get docker visible in docker ps print? Cause there's nothing for me
I have tried Devika, it works so smoothly like hot knife through butter. But yrr will try Open Devin soon.
Can it accomplish real task though?
Want to walk me through the setup and usage?
3:15 I got Devika working to create a simple "hello world" webpage with html and css files but it couldn't update or change anything once the files were created which was pretty annoying. Not sure if it was my fault of not understanding workflows. Looking forward to following this tutorial to try OpenDevin, thank!
What kind of machine do you need to run it locally ?Also what models should you use?
1) Don't mix pip and conda!! 2) Diagnose zsh/bash python alias problems with "zsh -x" 2) Try poetry? 3) Somebody share a working Docker image already setup! 4) Somebody compile all these python dependancies into a massive binary!
no way I'm using conda anyway.
For npm install is there a .nvmrc file that tells you what version of mode you need? You can try nvm use && npm i
That way the browser works better
So, what do you recommend? Pythagora or OpenDevin?
I'm waiting for a click and install app, this put you through the hoops for mediocre app development isn't worth it. It's very interesting though, i'll keep my eye on it.
Yeah it's not very userfriendly yet.
Shit is nuts! Let’s go!
Where is docker being used? I saw it was installed, but didn’t see where it was used.
Does Open Devin do jobs on Fiverr yet. Nice video good explanation.
Do you know by any chance where OpenDevin tries to run commands, is it in the workspace or in the docker container? Cause we got node, monogo etc. all installed and accessible for OD, but it keeps claiming it can't reach them, tries to install them, fails, quits. Can we make OD "work" on the regular bash, not inside the docker enviroment?
I don't need a tool that creates code from 1 description.
I just want the copy/paste automated. I just want an agent that if I sag Save it will automatically insert the code.
Is that so much to ask?
Looks good…until the initialing agent gets locked out 10secods.before the show on the MACbook -intel . Any thoughts.
So does it depend on the AI on what its allowd to write? Like how ethical itis or is devin controlling that?
kindly help about this python -m pip install -r requirements.txt , this file is not listed . also File "/Users/username/Downloads/Mine/Projects/OpenDevin/opendevin/server/listen.py", line 5, in
import litellm
ModuleNotFoundError: No module named 'litellm' , thanks in advance
Great video Matthew, I managed to get the backend and frontend but it is displaying "Initializing agent (may take up to 10 seconds)..." on the openDevin chat forever so I'm stuck. The only thing seems to be wrong in the backend is "Error creating controller." line. I'm running it using vscode on windows, any suggestions?
I got it to work on Windows. I also had the "Initializing agent (may take up to 10 seconds)..." but when I looked in the 1st terminal, I saw that it disconnected. It was because It did not pick up the API key. Once I got that fixed by setting it in config.toml then it worked.
You should use WSL it's a hard requirement now
I'm a little confused why the frontend and backend aren't just docker containers. Then the whole project could be run from a single docker-compose command.
Hold on. If it's dockerized, why do you have to mess with any installation packages anyway? GPT Pilot for instance is self-contained, requires no extra configuration (apart from API key of course) and just shows its console in a web browser.
I was able to get it started only to end up with a white screen on the front end 🤔
They should release it with a bash installation script in my opinion.
If you are using double quotes you need to use $HOME instead of the tilde.
Hi can you help with the android app development ai assistance run locally?
Devika is 100% free, runs on offline llms and builds apps too, never got it to do much more than minor stuff, that didnt required a bunch of debugging but fact remains, it will certainly make a calculator and other simple stuff like for sure. I just followed along with the indian chicks tutorial and it was installed and working pretty quick. I think I am going to go see if it will actually do anything worthy....
Any tips on getting good output with llama3?
is it me or y'all got an error with docker, do we need to run the container of the image that we pulled
At the moment there are so many projects appearing and becoming obsolete in no time that I'm struggling to decide which one to focus on and plan with for the longer term... it's a real pain!
C-z stop or suspend a command in the shell. It doesn't kill it.
Just try "jobs" in the shell to see that. Us "fg" to continue the command.
Hey matthew , whenever i try to use the opendevin after running it (done this 20 times at least), it doesnt load, keeps showing initalizing agent (may take up to 10 seconds), nothing else happening.
7 minutes in and I don't believe this is replacing developers anytime soon 😂
Exponential curve buddy. Exponential curve.
@@chineseducksauce9085 yeah. Once it happens devs won't know what hit them.
Yep... they said the same with arts and writing. Today special effects studios are closing and hollywood's writers are left with no jobs.
Just wait a little more and a lot of white collats jobs are going to disapear.
can it work with vision localy without paying for API tokens?
Hey there's a missing part here: How do we generate requirements.txt? I've tried pip freeze > requirements.txt but it doesn't seem to list all the required dependencies. Anything else you want me to try?
If you had to choose between any of the open source agentic apps, which would you pick? If I'm going to start with one, I'd like to stick with it. Which is best?