Research CodeAnt and Other Similar Tools and Draft Achitectural/Implementation Insights

[peachjelly13]

Research CodeAnt and Other Similar Tools and Draft Achitectural/Implementation Insights

Study tools like CodeAnt, SonarQube, and other similar platforms that achieve somewhat similar stuff.

CodeAnt AI - https://www.codeant.ai/ai-code-review

Things to do:

• Explore the features these tools offer
• Understand their architecture and workflow
• Identify what works well and what could be better (UX gaps, friction points, missing features)
• Think about how we can improve on their approach
• Propose our own flow of how things should work, combining the best parts and fixing the worst

End Goal: A clear architectural/flow document that reflects how our tool would work, inspired by but improved upon existing ones.

[lakshz]

I can take this.

[peachjelly13]

Sure, its an open discussion will assign once it starts out

[lakshz]

I tested CodeAntAI on a demo repo and reviewed SonarQube's docs. Here's the initial proposal:

📄 [Link]

Key Decision Needed

Self-hosted vs Cloud deployment

Both CodeAnt and SonarQube offer both options. For our MVP, we need to pick one:

• Self-hosted: Better privacy, more complex setup, docker-based deployment, no infra costs.
• Cloud: Easier to ship, but requires handling user code, we'll have to figure out infra costs

I've detailed the trade-offs in the doc. Thoughts on which direction we should take? @peachjelly13

[satyajitnayk]

We should use best model to generate critical(no BS) code review. One such explained in codeant.ai blog

[lakshz]

Yes, I gave thought about this too. Here's the trade-off I see:

Since we're open-source (not a business like CodeAnt), the API costs for premium models like OpenAI/Claude could be challenging initially. We can:

1. Start with open-source models via Hugging Face to keep costs minimal
2. Add support for users to bring their own API keys (OpenAI/Claude) if they want premium-quality

This way, open-source projects can use our tool with zero costs while still having the option for better models if needed. Let me know your thoughts on this.