51–60 of 140 posts
I wish there was a plugin / cli / TUI that would grab what you are doing and copy it into your clipboard for a chat app. That way I don't need to worry about tokens and usage limits.
You can install it in a few seconds, one line: curl 9ol.es/talker | sh
tell me what you think.
I've finally settled on this: - A custom vscode plugin to help me copy file contents from the tree view, along with file paths - A chat based ide (LibreChat, but anything will do) - An agent that syncs code back, once I'm happy with the conversation ( https://github.com/codespin-ai/codebox ) Sometimes I add the agent right at the beginning. Sometimes I do that when I feel the code is ready to be written back. Another…
How "easy" is it is to scaffold a personal vscode plugin?
However, I think it can be "vibe coded"these days.
I settled on gptel, which is an LLM package for Emacs. It's kind of a mix of both approaches: 1. You have chats right there in the editor, easy to copy/paste and manage context/history without switching to a browser. You can also quickly add files or portions of files to the context or remove them again. 2. You can choose which model you want to use for what, granted you have an API key. 3. You can quickly highlight…
have you tried aider? Using it with magit to see the changes is a good workflow. the aider integrations to emacs are also worth it, you can easily add/remove files and send highlighted parts of your buffer.
Related to this. ChatGPT, a lot of times provides “bare” diffs (ie. not complete unified diffs - they lack proper headers and line numbers) They are a pain to apply manually. I’ve tried several visual studio code extensions that apply patches from clipboard or provide an input for it, even saving them to a file and running command line tools - but they all fail Anyone has a good system, extension or application to ea…
I don't use IDEs to begin with. I use a text editor, since I am a UNIX man. So, I don't even have a reason to look at the "AI" flavor of an IDE. The closest thing to an AI IDE that I have tried out were tools like AIDER and Jack's "goose" agent. Neither of those specialized tools has been satisfactory. They all performed worse than just the LLM IMO. I am sticking to crafting my own context that I supply to the LLM. T…
Edit : Sarcastic
I've been using it for about a year and it's incredible.
My take is that most people did not invest the little time necessary to get acustomed to its workflow.
My advices for aider are:
- familiarize yourself with the chat modes (architect, code, ask)
- familiarize yourself with the edit modes (diff, whole, etc) and know which to use for a given model. Indeed not all models handle all modes equally well.
- make the code one feature at a time, by small chunks if needed, by limiting the contxt to the relevant files.
- practice to learn how to best phrase stuff.
- write you coding preferences into aider convention files. Things like "always use type hints, beartype for type checking, python click for the cli, add many comments".
I'm mainly doing python and with proper comments and type hints it's really easy for models to write code that works with the rest of the repo even with only a few files in its context.
I use Claude Code (cli tool) and it's on another level. Not even comparable to code autocomplete à la Copilot or copy/pasting into a LLM chat app. It knows about your whole code base, can use external tools, read documentation, run tests, etc. Even as an experienced developer, it's been a huge productivity boost. The main downside is that it can quickly get expensive.
Does it really know about the whole codebase? How big are we talking about? Last I tried, some time ago, llm didn't work very well "in the large". But I have never tried Claude Code.
I don't use IDEs to begin with. I use a text editor, since I am a UNIX man. So, I don't even have a reason to look at the "AI" flavor of an IDE. The closest thing to an AI IDE that I have tried out were tools like AIDER and Jack's "goose" agent. Neither of those specialized tools has been satisfactory. They all performed worse than just the LLM IMO. I am sticking to crafting my own context that I supply to the LLM. T…
I didn't know Unix users were not able to use IDEs. The more you know /s Edit : Sarcastic
VSCode/Cursor run natively under Linux.
Earlier quoted context omitted.
Does it really know about the whole codebase? How big are we talking about? Last I tried, some time ago, llm didn't work very well "in the large". But I have never tried Claude Code.
No it doesn’t, it tries to be smart about how it loads context that it thinks it needs to perform a task. In reality, in a large codebase I’ve found manual supplying the exact context to Aider with Anthropic’s models to work better.