I managed to connect a GitHub data source to an assistant. The settings seem fine but when I chat with the assistant, it does not seem to recognize / connect / search the connected data source.
Question:
Can a custom assistant access a GitHub Repository (and read the code)?
If yes, why it is not working on my end?
Thanks
Hi Pat,
By default Dust don’t synchronise code as most customers only want to view their issues / PR descriptions more than the code.
We’ve activated code synchronization on your workspace, can you try again?
It does not work. It does not answer my questions and keep saying it cannot access the repository. What else am I supposed to do to make it work?
Hello,
I can see that your code has been synchronized, so this is odd that it’s not using it.
Can you share your assistant’s instructions by any chance? (if they can be shared publicly)
I can also see the folders and the code itself.
Here is the prompt I am using.
Help me start my coding journey in Python.
I need advice and guidance to help me develop an AI API, build a robust architecture, write code, correct code or interpret code.
Each time you generate code, make sure you read first the latest version of my GitHub repository, named DocReview which is connected to this prompt.
I am using Replit as the IDE.
When you generate code, please respect the pyright-extended rules, especially this one:
Line too long (125 > 88)
Example:
Line too long
print(f"Unsupported file type: {content_type} for file {file.filename if file.filename else ‘unknown_file’}")
Correct
print(
f"Unsupported file type: {content_type} for file "
f"{file.filename if file.filename else ‘unknown_file’}"
)
Then, when I chat the assistant, I ask:
@cs explain me the orchestrator.py file.
Outcome: it invents a file which has nothing to do with the orchestrator.py file.
Hi,
Unfortunately this will not really work with Dust currently, as with the “Search” tool (which I guess you used?) we are doing semantic search, meaning that we will find parts of the code that can answer your query. So we cannot analyse a full file, so the LLM will start hallucinating.
We will release in the near future a tool called ‘include’ to be able to fully include files that will make your use case doable.
In the meantime, the best way for you do ask this is to attach the file using the paperclip icon in the input box
Thanks for your response and follow-up - although it is really a shame as everything seems to be in place (data source connected, code sync). Now I have to create static files or copy-paste code again and again which is not very convenient. Please let me know when there is update on that matter.
Will do - you can keep an eye on Latest updates to see the latest updates.
Being able to include entire files from datasource is definitely on our roadmap, just not with a visible timeframe for now!