Side question : I test the chat and ask him to do a PR for my issue 108. He looked the issue and write a typescript code whereas my whole project is in dotnet.

• Is it context aware ? He can read the code of my project

or maybe you can add a "+" button to add files from the repo

that way we can choose which file we want to send.

And since I don't know what he is doing of if he ever doing something or it crashes...

AI user will always post its result back as comment, including potential errors (timeout, permission issue etc).

Is it context aware ? He can read the code of my project

When review a pull request, it has tools accessing source code.

Maybe adding an immediate comment saying something like "AI is thinking..." / "Please wait, I'll take a look..." and then updating that comment (or adding another comment) once AI completes or fails would solve the feedback issue mentioned above.

Robin,

I'm a big fan or OneDev for many years and you are doing awesome work on it.

But I tested the AI feature and nothing is really working, it is just a chatgpt frontend and nothing is context aware.

Example :

In French, I ask him to create a build job for my dotnet application, which very obviously is in the repo and the only one in that repo. It could not be wrong. And we are in OneDev, so I expect it create a valid onedevbuildspec.

Its responses :

I need more info : What build system do you use ? What version of .NET What type of project...

It should not ask all this question if really context aware and I have many other examples of this.

Maybe I'm doing it wrong, you could show real world example, it could help.

Side note : I use ChatGPT for 3-4 months from now in my today workflow to speed development. I use it as coworker and I know how to handle/iterate though its work. The only advantage to have it in OneDev is the context aware !

For me, context aware is :

• I am in OneDev so everything that I ask about CI/CD should related to how OneDev do it (build, job...)
• When I ask to code something, it should use my current codebase's project (the simple fact he could access the list of files, could help him to look at the interested one, ex: .csproj for dotnet to get the NET version). I understand for performance/cost reason it can read all my code, that's why I can help to upload or name the file he could use for its thinking.
• Instead of writing code in the chat, he could write file in a branch of the current repo. So then, I can accept/review its PR.

I hope I could help to improve the AI in OneDev

EDIT

If I explicitly name the project, it find it, and I see some files uploaded. Sorry for my previous comment, I think I was doing it wrong, the prompt was lacking info. Maybe an example or a tip for user to name project in the prompt could help.

Context aware means that:

1. If you want to add a job to build spec, you first need to enter into build spec edit page
2. If you want to summarize comments of an issue, you first need to switch to that issue
3. If you want to ask questions about codebase, you first need to enter into the files page of a branch (or tag/commit)
4. If you want to ask questions about a diff, you first need to open the relevant diff.

In chat mode, AI user uses your own identity, so you need to open any resource for it to work on.

Having AI user work on a separate branch and send pull request when finished is also considered in future versions, possibly by assigning issues to an AI user.

Having AI user work on a separate branch and send pull request when finished is also considered in future versions, possibly by assigning issues to an AI user.

That would definitely be a banger !! And a very efficient way to work with AI.