> I need something that can actually look at my whole folder and understand how the different files talk to each other Oh man, I totally get the frustration! If you want to save some cash but still get that deep context, you HAVE to check out Continue Dev Open Source VS Code Extension. It is amazing because it is free and lets you plug in whatever API keys you want. Since you have a budget, just grab an API key for OpenAI GPT-4o API or even better, the DeepSeek Coder V2 API which is ridiculously cheap and punches way above its weight for refactoring. It basically gives you that whole folder context without paying a massive monthly subscription fee. Also, if you are doing boilerplate, Tabnine Pro AI Coding Assistant is fantastic for local stuff and they usually have some decent trials. Honestly, mixing Continue with a cheap API key is the best way to keep your costs down while getting elite-level refactors! You basically pay only for what you use, so if you are grinding hard for 12 days, it will still be way under your $50 budget. Love it when a plan comes together like that!

Jumping in here... honestly, I was in the same boat a few months ago with a messy legacy project. I was using some random free extensions and they were just... bad. I finally bit the bullet and got GitHub Copilot for Individuals and it made a massive difference for my workflow. The key for me was actually using the chat feature with the @workspace tag. Before I figured that out, it felt like the AI was just guessing based on the current file, which is totally useless for spaghetti code. It is a pretty decent option because it actually indexes your local files properly. I used it for a few specific things:

• tracing variable origins across different modules
• generating boilerplate unit tests for old controller files
• creating JSDoc comments for functions that had zero documentation It is not perfect—sometimes it gets a bit confused if your file structure is really messy—but it is way more reliable than the basic autocomplete. For 10 dollars a month, it is a steal and fits your budget perfectly. I used it to generate about 40 unit tests last month and while I had to fix some logic errors, it saved me a ton of manual typing. Just a heads up on the safety side tho... always check the database queries it suggests. Legacy systems sometimes have weird constraints that the AI might miss even with the context turned on. But overall, it has been a total lifesaver for me. You have got this!

Honestly, check YouTube! Theres a fantastic video analyzing these specific workflows for legacy projects. Just search for AI developer benchmarks... its really the most reliable comparison.

Regarding what #4 said about "Jumping in here... honestly, I was in the..." - i agree that some tools are unfortunately not as good as expected when you look at the market analysis. I had issues with their reliability during a recent audit. It reminds me of a trip to Ohio last year.

• my rental car broke down
• spent hours in a diner
• missed the keynote I spent the weekend arguing with a mechanic about spark plugs instead of networking. Anyway lol sorry kinda went off topic there.

Yo! Jumping in here because I am obsessed with legacy refactors! Like someone mentioned, context is everything, but I gotta ask—what testing framework are you forced to use for this 2016 mess? Knowing if its Mocha or something older helps a ton because the boilerplate needs to be specific. If you want long-term sanity, you seriously need to look at JetBrains WebStorm combined with JetBrains AI Assistant. Here is why it is amazing for your specific nightmare:

• Deep indexing: Unlike VS Code extensions, it builds a massive index of your entire dependency graph and local symbols before the AI even touches it.
• Semantic refactoring: When you rename a variable in a 2016 spaghetti file, it actually updates the references across the whole project without breaking things.
• AI Context: The AI Assistant uses that deep IDE knowledge to generate JSDoc that actually reflects reality, not just guessing based on the file name. Honestly, for complex Node stuff, the static analysis in JetBrains is a lifesaver for ownership because it finds bugs before you even run the code. Its a total game changer for these messy client projects! You can definitely fit the sub into your $50 budget too. Let me know about the test runner and I can give you some specific prompt tips for it!

Man, I totally feel that laptop-out-the-window energy. Legacy Node stuff is a total nightmare when there is zero documentation. If you're worried about reliability, you might want to consider switching to Cursor AI Code Editor Pro Plan for a month. Yeah, its a fork of VS Code, but the way it indexes your whole folder is way more reliable than the standard extensions for understanding how files talk to each other. Just be careful tho, id suggest keeping these things in mind:

• Start with Sourcegraph Cody Enterprise Individual if you need deep context. It reads your local files better than most things ive tried.
• Use Anthropic Claude 3.5 Sonnet for the heavy logic. It is way less likely to hallucinate than older models.
• Make sure to manually review any refactors it does to your database calls. Legacy schemas can trip it up. Be sure to check the privacy settings first so your clients code stays safe. You can do this!

Nice, didn't know that