Anyone using AI for code reviews with DeepSeek?

Hey everyone,

I’m looking for an open-source tool that can automate code reviews on GitHub pull requests using AI. My must-haves:

  • Free and open-source
  • Lets me use my own API key for different models
  • Ideally works with DeepSeek models

I built a custom GitHub Action using GPT-4 for reviews, and while it worked well, the token costs were ridiculous, especially for big repos. With DeepSeek’s cheaper pricing, I want to give this another shot.

Has anyone come across a tool that fits these requirements? I did some searching but haven’t found anything solid yet.

Any suggestions would be great!

DeepSeek R1 isn’t really great for coding. You might want to check out some fine-tuned Llama models instead.

Ellis said:
DeepSeek R1 isn’t really great for coding. You might want to check out some fine-tuned Llama models instead.

V3 is better for coding, R1 is more for planning.

Ellis said:
DeepSeek R1 isn’t really great for coding. You might want to check out some fine-tuned Llama models instead.

Are you talking about DeepSeek R1 Distill Llama 70B?

Sadie said:

Ellis said:
DeepSeek R1 isn’t really great for coding. You might want to check out some fine-tuned Llama models instead.

Are you talking about DeepSeek R1 Distill Llama 70B?

Yeah, that one.

I found something similar for GitLab. It’s a PowerShell script, but I converted it to Bash with AI, so you might be able to adapt it for GitHub too.

https://medium.com/@kedargmnv/automating-code-review-with-openai-in-your-gitlab-workflow-c18b894bf632

Isn’t it usually the other way around? AI writes code, and humans review it. Why flip that? So devs do all the work, and AI just reviews? AI isn’t better than a real developer.

Rayne said:
Isn’t it usually the other way around? AI writes code, and humans review it. Why flip that? So devs do all the work, and AI just reviews? AI isn’t better than a real developer.

Or maybe AI writes and AI reviews? :joy:

For solo projects, I write all the code myself, but I still use pull requests because they help me catch mistakes. AI code review would just be an extra set of ‘eyes’ to spot issues early. I’d still manually review everything myself, but having AI as a backup doesn’t hurt.

Rayne said:
Isn’t it usually the other way around? AI writes code, and humans review it. Why flip that? So devs do all the work, and AI just reviews? AI isn’t better than a real developer.

If AI can write code, it can review it too. It doesn’t have to be better than a human—just another tool to catch things you might miss.

@Hartley
If I don’t know something, I don’t code it—I ask AI to do it for me first. :sunglasses:

Rayne said:
@Hartley
If I don’t know something, I don’t code it—I ask AI to do it for me first. :sunglasses:

And you probably learn something in the process. :slightly_smiling_face:

I got better results with DeepSeek for coding than I did with Google Gemini. Haven’t tested it against GPT or Llama though.

I actually built something like this! I posted about it, but it hasn’t gotten much attention yet.

https://www.reddit.com/r/SideProject/comments/1i9y87f/i_built_an_aipowered_pr_reviewer_that_chats_with/

It only works with OpenAI models for now, but I’m adding Anthropic soon and could probably add DeepSeek too. Let me know what you think!

@Hartley
Nice, I’ll check it out!

Jin said:
@Hartley
Nice, I’ll check it out!

I just added DeepSeek support to a new branch if you want to try it out:
https://github.com/uncazzy/pr-reviewer/tree/feature/deepseek-integration

Still need to test it more, but should be ready later this week.

Has anyone used Qodo for code reviews? I haven’t tried it, but it looks interesting. I think it supports multiple AI models, but I’m not sure if DeepSeek is one of them.

https://github.com/qodo-ai/pr-agent