#First code assistant in the world to apply multi-file code changes (like a next-gen GPT-Engineer)

113 messages · Page 1 of 1 (latest)

weary obsidian
#

This isn't a replacement for Github Copilot, think of it like a different flavor altogether. With this you can create entire codebases from scratch, work on an existing codebase, talk about your code with the AI, or get the AI to write code/documentation/tests for you. All powered by GPT4, and completely free while we are looking for feedback.

Some examples of what you can do with Codebuddy:

Ask it to implement a new feature, without being specific
With the current version you still need to manually select which files to include in the context, but when you do you can ask it to make a multi-layered change. For example, let's say we want to get a new user's birthday. You can select all the files it would need to make the changes to and then it can make those changes to each file (service, controller, UI..etc). If you're not sure what files to include, talk to it first!

Create new files and use existing files as a reference
Sometimes you want to quickly create a new command or UI component and you have a couple good code references in your project already that you want to use. Just select those files and ask it to use those references to create the new component.

Encourage the AI to ask you any questions if something is unclear
The AI will sometimes ask you followup questions when something is obviously unclear, but you can also prompt it to do so specifically if you want to hash it all out with the AI first before implementation.

Always know how much context space you have
Codebuddy features a dial that shows you how much context space you have left when including files, long prompts, or other inclusions.

Use voice input to further speed up development
Since Codebuddy often feels like a junior developer, it can feel a bit slow to have to type out everything you want to say. Hold the right-alt for PTT or press the mic button to talk.

It's free! It's GPT4! Check it out and we need your feedback!
https://codebuddy.ca

lost lava
#

it should help your downloads

so many people wont download without seeing how's it

weary obsidian
#

I'm working on some demonstrator gifs to show what it looks like. But yea, good idea

lost lava
weary obsidian
#

I'm putting a couple together right now then, yea

#

An example prompt that would take multiple files, just asking the AI what files I'd likely need to include in my prompt to add a birthday to the user signup experience. It has gleaned the filenames quite accurately despite only having the filenames to work from (future versions of codebuddy will also have file summaries in various levels of detail so it can give you better answers and select files for you).
https://i.imgur.com/AvgTger.png

#

Changes are being applied automatically
When you make a change it will automatically apply those changes to your files, then you can look at the DIFF to confirm that the changes it made were good and the application was successful.
https://i.imgur.com/pKdzqbR.png

#

Here is the diff view with the changes it has given me so far
It ended up changing both files in this case, but you can also simply include files as references that aren't meant to be changed. If you don't have enough context space to make the changes to ALL the files at once, you can do it piecemeal, and use the reply button to continue in the conversation where you left off. Replying within a conversation will include the chat history as part of the prompt. New conversations allow you to start a fresh conversation and clean up the context history.
https://i.imgur.com/Hz4hoTz.png

#

Pushing your changes, it auto-generates the commit message based on your prompts
You can edit the commit message before pushing it if necessary, but it'll autogenerate and tend to be usually 1-2 sentences, sometimes 3.
https://i.imgur.com/J3OPACm.png

#

@lost lava Thanks for the suggestion!

lost lava
#

Nice!

You should add a email-signup version.

Because in your github auth, it needs acess to all the private repositories

It makes so much people exiting of your app, because they wont give acess to their private repos

weary obsidian
#

hm interesting thought, yea...

#

currently we only support github so you'd still need to authenticate through it, but maybe we could have a version of authentication that doesn't include private repos.

#

I actually do have an email signup version in the code including UI and all, I just disabled it because I don't have a codepath for signing up and then also authorizing github via a u/p signup.

#

also by forcing a github account requirement, hopefully I can avoid getting people trying to use this for non-coding purposes a little ^^

#

@lost lava I like the suggestion though, I'm putting that on my board

#

thanks!

lost lava
#

also works

#

or change your github oauth permission to just "public" repos.

Because private repos are so confidencial

weary obsidian
#

most people have a google account though, it wouldn't do much to prevent non-dev types

weary obsidian
#

I'll have to do that I see that now

lean rover
#

does it also interpret code?

weary obsidian
# lean rover does it also interpret code?

It can't run it, no, however you could setup github actions to automatically run it when you push.

The devcycle for the web version is to push to your branch, pull it in your dev environment and check it out... then repeat

#

I do hope to add integration with github actions so it could at least compile and run tests and try to fix issues for you

#

but that's not in this version yet

#

If you already use ChatGPT for coding then this is a clear winning replacement for that at least imo

#

in fact, most of Codebuddy was written using ChatGPT at first, then using codebuddy itself

weary obsidian
#

During this phase, you're using our API key

mighty otter
#

like u had to talk with openai or

weary obsidian
#

oh, no you can get it from the openai website

#

it's easy to generate one

mighty otter
weary obsidian
#

huh. i thought gpt4 was generally available now... it should be anyway

#

Try creating a new account perhaps?

mighty otter
#

and look

mighty otter
weary obsidian
#

you know what

mighty otter
#

idk

weary obsidian
#

I saw a random post here today where someone said they didn't have gpt4 access, so they revoked their billing account and then redid it and they instantly got gpt4 support

#

i'm thinking creating a new account will work too

weary obsidian
#

Trying to find it but I'm not sure... it was in one of the channels

#

maybe within the last 3 hours

mighty otter
#

If I put the amount like to $2 dollars, and they charge me, does that mean I get instant access?

#

This is what I mean

weary obsidian
#

not sure...

#

Personally I got GPT4 access by providing an eval in their open source eval project (used for training the models). But since then they released it publicly, I think they just didn't insta-migrate everyone to it and it's done for new accounts automatically as part of the signup process.

#

But the method that other guy mentioned might work for you... not sure if you have a charge pending though..

fallow halo
lost lava
lost lava
fallow halo
lost lava
#

i think it was added 1 year ago

weary obsidian
covert spear
#

Nice work man!

I created a smaller but similar-ish project for code reviews/explanations using chrome extension, I went that route because I know some people are afraid of giving GitHub auth

Feel free to look into it! https://codesensai.com

weary obsidian
#

Hey I've seen yours in the search results. Good job man! 👍 @covert spear

#

I updated the Codebuddy website with new screenshots and a new look on the landing page (best viewed on desktop):
https://codebuddy.ca

cunning cape
#

Hi @weary obsidian . Looks like a really cool project. One question though. Why does codebuddy need this level of access? Writing access to personal user data seems excessive?

weary obsidian
cunning cape
#

Thanks @weary obsidian , I would love to test the product. It looks pretty amazing 😉

weary obsidian
#

Awesome! Can't wait to hear about everyone's experience

#

Good or bad, it's all invaluable

weary obsidian
#

First AI assistant in the world to apply multi-file code changes (like a next-gen GPT-Engineer)

weary obsidian
#

First AI assistant in the world to apply multi-file code changes (like a next-gen GPT-Engineer)

#

Here is a live video of Codebuddy in action, creating a new feature in an existing web application that spans multiple files:
https://youtu.be/XmHjwKQ12KA

This video shows how Codebuddy is able to make multi-file changes at once, shows how you can use conversation to plan out the changes just like you would with a real person, APPLY those changes directly to your codebase, and then check the changes with the DIFF viewer before pushing to your branch.

▶ Play video
spring lily
#

Excellent - I've been using it to correct some of my coding errors on projects I am working on and it's been amazing. This truly is a game changer!

fiery thorn
weary obsidian
#

Thanks so much!

pearl glen
#

+1. I like this approach and tests, keep it up!

crisp lotus
#

Hey @weary obsidian based on the videos this looks awesome.
I am having an issue though, or maybe I'm missing something. I logged in with github on the the codebuddy web version, it recognised all my repos correctly, but when trying to open one, I see no files, and I can't for the life of me find any solution

#

Tried with both public and private repos, I even tried to make a new one within codebuddy, and it showed up correctly on github. I added files on github but they still won't show on codebuddy

#

I also have other questions about the future of this project, but I'd like to be able to try it first lol
(Please ping me if and when you reply GioYay )

weary obsidian
#

@crisp lotus if you are using Safari that might be the problem. There are some issues that need to be addressed still to make Safari work. Try using Chrome

#

(I blame Apple for moving away from V8)

kind orchid
#

If this is working, that’s a game changer! Would love to try!

jovial peak
#

Does it works like onboard.ai too? Like asking the AI a question and receive answers based on the workspace or github repo files?

weary obsidian
weary obsidian
kind orchid
weary obsidian
kind orchid
#

Nice! Coding is already massively faster now with copilot and gpt-4. But a lot of copying and pasting going on 😄

weary obsidian
#

YES, I couldn't stand that part

#

and now there's voice integration and website URL inclusions

#

Soon Iron Man <-> Jarvis imo

#

Soon™️

#

^^

kind orchid
#

It’s unreal the innovation happening 😂

weary obsidian
#

oh speaking of copilot

#

copilot tends to have weaker solutions too because of the lack of chain-of-reasoning (imo)

#

Codebuddy starts formulating a plan with you and waits for you to OK it before writing code

#

that seems to make a big difference, plus you can review the plan to confirm a lot easier before it's turned into code

#

once it's coded though, the AI tends to not want to change as much

#

so it's better to do that while it's still prose

weary obsidian
weary obsidian
#

Nice! I'd love to hear your feedback if you get a chance

weary obsidian
#

Update

You can now use voice in your IDE! It's sooo much nicer to just talk about what you need to have done, and I'm pretty sure the code quality is better for it because AI likes when you're more verbose (which you naturally tend to be as you're describing what you need).

https://codebuddy.ca

weary obsidian
#

Update 2

We just started testing the VS Code plugin! Until now we've only operated in the browser, and on Jetbrains IDEs, but soon you'll be able to use Codebuddy right inside of VS Code. We'll make an announcement to everyone that has signed up when it's released probably within the next week or two.

https://codebuddy.ca

weary obsidian
#

First code assistant in the world to apply multi-file code changes (like a next-gen GPT-Engineer)