Aider

“After using Aider for a few weeks, going back to co-pilot, roo code, augment, etc, feels like crawling in comparison. Aider + the Gemini family works SO UNBELIEVABLY FAST. I can request and generate 3 versions of my new feature faster in Aider (and for 1/10th the token cost) than it takes to make one change with Roo Code. And the quality, even with the same models, is higher in Aider. Anybody else have a similar experience with Aider? Or was it negative for some reason?”

“So, I have been trying to use aider as a pair programmer tool with Qwen3 models, but it is just a disaster. Editing files without asking for permission, creating new duplicate folders/files... it just mess with the whole project. Does anyone have an open-source alternative to it?”

“Use it every day and sing it's praises to all who will listen, this tool generates 90% of my Sonnet and O3 traffic the only real downside is strong/fast models cost a few dollars a day. For professional devs it's a no brainer.”

“Aider clicked for me once I realised it commits each change as its own git commit, so the model's work is reviewable and revertable like any teammate's. Being model-agnostic means I can point it at whatever is cheapest or strongest that week. It rewards discipline: feed it the right files with /add or it burns tokens guessing. For terminal-first, git-native AI coding it is hard to beat.”

“--watch-files parameter saved the tool for me. It feels more natural to work in the code then when I need AI, I inline a comment and aider will write the rest. e.g. I don't have to lift my hand to use the mouse, highlight some code, right click to popup a chatbot menu and ask the LLM--I just inline the comment, save the file and aider will do the rest. # Why isn't this working, fix it for me ai! def add_two_numbers(a, b): return a - b And aider will use the LLM to fix the code. Caveat: if you do”

“I am experimenting with both Cursor and Aider and find them great, but it is difficult to review changes coming from Aider. I came up with the following solution, it is a little bit of work, but utilizes the traditional PR review mechanics. If you have your project hosted ex. on Github, create a feature branch, let the aider do the work with the auto-commits on, then push the branch and create a PR. You can easily review the diff similar to what cursor does, and let the aider follow up on any ne”

“I strongly prefer aider exactly because it's not tied to an editor. Having to switch editors to use an LLM for coding isn't a serious option. And it's not tied to an LLM either. Gemini overtaking Claude? Just a pref flip away. Compare this to having to switch your entire IDE around. LLM stuff isn't mature enough for this kind of commitment.”

“Aider clicked for me once I realised it commits each change as its own git commit, so the model's work is reviewable and revertable like any teammate's. Being model-agnostic means I can point it at whatever is cheapest or strongest that week. It rewards discipline: feed it the right files with /add or it burns tokens guessing. For terminal-first, git-native AI coding it is hard to beat.”