Hey everyone! I use AI to review my diff files for pull requests. My goal is to review each hunk of the diff file, so I want the review from AI to be specific to each hunk. Currently, I loop over the hunks, and for each one, I make a request and then save the feedback with the hunk data. This approach is costly. I’m looking for a solution to reduce the number of AI requests while still getting feedback from the AI that is specific to each hunk.

Hi :) I just want to drop 2 of my recent Videos, maybe some find it interesting. --- Find Context in Large Codebases with /read-only-stub (aider) I have created a Draft PR for aider: The new command /read-only-stub allows the inclusion of read-only files as stubs in the chat context. This feature lets developers add file outlines without embedding the entire content, helping keep the context clean and manageable. You can use whole repos or a folder as context for initially picking which files are relevant for a user query. For further investigation the relevant stubs can be expanded. Similar to the repo-map feature, but more explicit and deterministic. --- MCP Servers in Aider: A Practical Experiment I’m showing a cool experiment where I connect MCP servers to Aider using a CLI tool called mpcm-aider. Although Aider doesn’t natively support MCP, this tool acts as a proxy to pull in external context like AI developments and Obsidian file contents. I walk through installing mpcm-aider from NPM, configuring MCP servers like Perplexity and Obsidian, and using them seamlessly with Aider. Follow along if you’re into generative AI and want to see more fun experiments. Happy coding!