Ask HN: GitHub Copilot Alternatives
I'm currently using GitHub Copilot, but I often see problems interpreting commands and limitations in responses when compared to more advanced models (Claude/ChatGTP 4).
I've searched a few times, but I've never found any references to the implemented version (model).
So, I ask: what development tools are you using? Copilot, or are there better alternatives?
I use it in conjunction with IntellIJ. If you're just looking for something that's similar in the auto-complete category, I'd suggest adding the Continue plugin/extension (https://docs.continue.dev/quickstart#jetbrains) and pointing it at Claude 3.5. If you haven't tried it yet I'd also suggest trying https://aider.chat/. It's outside of the IDE but I find it's the right level of 'agent' for where we currently are with models. Even if you use it for nothing else it's nice to be able to run the `/test` command and have it use the test output to fix errors. Interesting. Claude has (at least in my tests) best results than GitHub Copilot. About using in IntelliJ, I read some people talking about changing the Java Boot. Do you know what configuration I need to change? "Keep in mind you currently may have to change your Java Boot Runtime for the IDE to make it work." Here's a version that keeps closer to the original text while fixing the main typos: Very happy with https://supermaven.com. Switched to it from Copilot, and it's faster and has a much better understanding of what I'm trying to do. I think it uses a git diff + some other stuff that it feeds into the context window that makes it so that when I'm working on a new feature, when i’m creating say the 3rd it understands what I'm going to write even if the file is empty. Supermaven is brilliant, been using it for probably 4 months and can't imagine ever cancelling this subscription unless something better comes out. Supermaven didn't work for me at all. Tried contacting support - no reply. On the other hand, TabNine serves me well for several years now. I'm working on a CLI tool which helps to generate and modify a codebase: https://github.com/gtanczyk/genaicode/ It is still a WIP, and pending a major overhaul (both implementation and the UX), but it works already, and helps me realise my ideas (games): - https://www.gamedev.pl/games/monster-steps/ - source: https://github.com/gamedevpl/www.gamedev.pl/tree/master/game... (the idea and majority of code is AI generated) - https://www.gamedev.pl/games/nukes/ - source: https://github.com/gamedevpl/www.gamedev.pl/tree/gh-pages/ga... (my idea for game, a lot of code is generated) also the tool is developed using the tool itself the difference compared to main-stream tools is that I can adjust it to my needs as much as I want. I've had a great experience using the IntelliJ plugin called CodeGPT (https://plugins.jetbrains.com/plugin/21056-codegpt). What I really like about it is that it allows you to bring your own API key for various providers like Anthropic and OpenAI. Additionally, CodeGPT supports local LLMs, which can be useful if you're concerned about data privacy or want to work offline. Seems to be interesting. Unfortunately, I can't run anything local with reasonable time response. By the way, I need to search again about this, maybe de Llama local runs better now with recent improvements on the model. Thanks by sharing your experience. I use Tabnine - https://www.tabnine.com/ i created directory for such ai agents and i have a category for coding:
- CursorAI
- Continue there are many others pen source and close source
https://aiagentsdirectory.com/category/coding Cursor [1] has excellent codebase RAG, and great recall of third party documentation. Use "instruction @web @Flask" to search the internet and Flask docs before answering, for example. I think Cursor is very useful, fully compatible with vscode, basically do not use their own code, write their own instructions, Cursor is fully executed, no matter how many times to modify, it will not complain. You can try it. I think you'll love it.
Cursor: https://www.cursor.com/ Any alternative is good as long as you don't care about the code. If you care, a generic autocomplete is enough. I have been using Zed for the last several months, and their new Zed AI addition is quite good, which I think is only available with their Preview (i.e. beta channel) release this far. It also comes with free in-app access to Claude 3.5 Sonnet (at least for now). CodeGPT is one of the most versatile, offering nearly the same features as Copilot. It also provides portable AI agents with context awareness and can be used in VSCODE, IntelliJ, and Cursor. Sourcegraph Cody (https://sourcegraph.com/cody) FOSS, self hosted: https://tabby.tabbyml.com/ At this point there are so many good alternatives to Codepilot! I've been using codeium and happy with it, mostly cause it's free... I'd like something with the addition of vim-supported. I just want to have a shortcut to see a suggestion I've been using codeium in vscode I'm hearing some good things about Cursor, it's a fork of VS Code with AI built in, subscription based. Some VSCode extensions from a quick search: Tabnine Amazon Q Continue Codeium Blackbox CodeGPT Gemini Code Assist Cody Double With all these agent, I am beginning to feel they all have the 'same' source and only differentiate in the UI/UX component. My mind is into the route to use AI agent as an exercise to create yet another, AI agent. I tried cursor recently. It has potential but as it is is very alpha feeling and clunky. Better than copilot but maybe less productive. And probably worse than the copilot, aider and vscode combo In some ways it makes sense to run your own LLM locally, one that's optimized for your tech stack and code base. For this reason I prefer the concept of Continue to Cursor. Cursor is good, but it can't use with IntellIJ vscode has continue. I heard that it also has an intellij plugin which sucks.