As a developer, I’m always looking for tools I actually control.
Stuff I can run on my own machine, mess with if needed, and not think twice about what I’m sending into it. I’ve used a lot of AI tools built for devs. Design tools like Claude Design, database managers, CLI tools, all that.
They work. But yeah… there’s always something. Either you’re keeping an eye on tokens, or you hesitate before dropping real project data into it. I’ve caught myself doing that more than once.
So I started trying open source alternatives. Just wanted to see what’s out there. A few of them stayed. I'm Still using them.
Some made things faster. Some just removed annoying setup steps I didn’t question before. And a couple I use just because it feels better knowing everything stays local. These are the ones.
1. LLamaFile
Running a local LLM usually starts the same way.
You install Python. Then something breaks. Then CUDA. Then something else breaks. At some point you’ve got three tabs open, one half-working environment, and you haven’t even run a single prompt yet.
llamafile skips all of that.
You download one file. That’s it.
On Windows, you rename it to .exe. On Mac or Linux, you run chmod +x. Double click or run it, and suddenly there’s a local AI model running on your machine. Open your browser, go to http://127.0.0.1:8080, and you’ve got a chat interface ready.
There are two ways to use it.
The easy way is grabbing a prebuilt .llamafile where the model is already there. One file, everything inside it. You run it.
If you want to try it, I wrote a quick setup + model guide here.
2. Dyad
I didn’t expect to like this one as much as I did.
Most AI app builders feel impressive for five minutes. You generate something, it looks decent, then you try to actually use it… and things start falling apart. Either you hit limits, or you realize you don’t really control what’s going on.
Dyad runs locally, which already removes half the hesitation. You’re not sending your project somewhere random, and you’re not tied to one model or one platform. You can plug in whatever you want. OpenAI, Anthropic, local models, your own setup. It doesn’t really care.
What surprised me more is that it’s not just a UI generator.
You can build full stack apps. Database, auth, backend logic, all wired up. It even hooks into stuff like Supabase so you’re not starting from zero every time. And if you already have something, you can import it, tweak it, export it again. No weird lock-in feeling.
There’s also this balance it hits where you don’t feel lost.
Some tools give you too much freedom and you end up staring at a blank screen. Others over-control everything and you’re just clicking buttons. Dyad sits somewhere in between. Enough structure to move fast, enough freedom to not feel boxed in.
Is it perfect? Not really.
You’ll run into rough edges. Sometimes things don’t behave exactly how you expect. But it’s one of those tools where you can see the direction clearly. And more importantly, you can work around things because you actually have access to what’s going on.
I ended up building a couple small apps with it just to test things out… and didn’t feel the need to switch tools midway. That rarely happens.
If you’re someone who likes building locally and hates the idea of being tied to one platform, this is worth trying.
3. HoppScotch
If you’re still using heavy API clients and waiting for them to boot up… you’re missing out on a masterpiece.
Hoppscotch is what happens when someone actually cares about speed.
It opens instantly. No login wall shoved in your face. No “workspace setup” ritual. You paste an endpoint, hit send, and you’re already looking at the response.
But it’s not just fast for the sake of it. It’s surprisingly complete:
REST, GraphQL, WebSockets, even MQTT if you’re into that
Pre-request scripts and test scripts (yes, you can automate stuff)
Collections and environments that don’t feel like a chore to manage
Works as a web app, desktop app, or even self-hosted if you want full control
And the UI… this is where it quietly wins. Clean, minimal, nothing fighting for your attention. You don’t feel lost inside it.
One thing I personally like: it doesn’t try to act like your entire backend platform. It stays an API client. That sounds obvious, but a lot of tools forget this and become bloated.
If you rely heavily on deep enterprise features or team-heavy workflows, you might still lean toward heavier tools. But for solo devs or small teams, this hits a sweet spot.
If you want to try it, I’ve put a quick guide and setup here
4. Codex CLI
Codex CLI sits in your terminal, where you’re already working, and just… stays out of the way.
You describe what you want, and it writes code that actually fits your project. It picks up your structure, your naming, the way your repo is already wired.
When you’re dropped into a messy codebase. You point it at a file or a folder and ask what’s going on. It reads through it and explains things in plain terms. Not perfect, but good enough to save you from digging through 20 files just to understand one flow.
I’ve also used it for quick reviews. Things like “does this break on edge cases?” or “am I missing something obvious here?” It catches small stuff more often. Not everything, but enough to make it part of my workflow.
Debugging is similar. When something breaks, instead of manually tracing everything, I just ask it to follow the issue. Sometimes it nails it. Sometimes it doesn’t.
The real win though is automation. Refactors, setup steps, repetitive edits across files. The kind of work you know how to do but don’t want to spend time on. It handles those as well
One thing to keep in mind that it’s not local-first like some of the other tools here. You’re still relying on external models, so privacy and cost depend on how you use it.
5. Velo
I didn’t plan on adding an email client to this list. But this one kept pulling me back.
Most inbox tools either lock you into their system or try to do too much. Velo takes a different route. It runs locally and keeps your emails stored on your machine using a simple SQLite setup. No weird middle layer, no “your data is somewhere in the cloud” feeling.
The interface is built for speed. Keyboard first, quick navigation, and search that actually understands what you’re looking for. You can use Gmail-style queries or just type normally and it figures it out.
It also has AI baked in, but not in an annoying way. You can pick which model you want to use for summaries, replies, or drafts. Claude, GPT, Gemini, whatever you prefer. Or don’t use it at all. It doesn’t force anything on you.
A few things that stood out while using it:
Emails get auto sorted into categories like primary, updates, and newsletters, and it’s surprisingly accurate
You can chain actions together, so one click can archive, label, and reply at once
Works with pretty much any email provider, not just Gmail
You can snooze threads, schedule emails, and even undo send if you mess up
There’s built-in phishing detection, which caught a couple sketchy mails for me
It also bundles newsletters together so they don’t clutter your inbox all day. Small thing, but it makes a difference.
Is it trying to replace everything? Not really. It just makes email feel less annoying, which is honestly enough.
If you prefer having control over your data and want something faster than typical web inboxes, this is worth a try.
6. OpenDesign
OpenDesign feels closer to a real design workspace than most AI tools in this space. You can generate layouts, tweak components, adjust spacing, and actually iterate instead of starting over every time. That alone makes it usable.
What I like is that it doesn’t lock you into one way of working. You can treat it like a generator when you need quick ideas, or slow down and refine things properly. It doesn’t fight you either way.
It’s also open source, which changes the equation a bit. You’re not stuck wondering what’s happening behind the scenes or whether your designs are being used somewhere else. You can inspect it, run it, modify it if you care enough.
Not everything is perfect. Some outputs still need cleanup. And if you’re used to very polished design tools, you’ll notice gaps. But it’s good enough to actually use, not just experiment with.
I’ve used it mostly for quick UI ideas and rough layouts before jumping into final polish elsewhere. It saves time, especially when you don’t want to start from a blank canvas.
If you’re curious about AI in design but don’t want to rely entirely on closed tools, this is a solid place to start.
7. GitButler
I didn’t think I needed another Git tool either. Git already works… until it doesn’t.
You know the moments. You’re juggling branches, rebasing, fixing commits, trying not to break history. It works, but it’s not exactly fun.
GitButler changes that experience in a pretty noticeable way.
The biggest shift is how it handles branches. Instead of constantly switching back and forth, you can work on multiple changes at the same time. They stay organized, and you don’t lose track of what belongs where. It sounds small, but it removes a lot of mental overhead.
Editing commits is also way less painful. You can tweak, reorder, or split them without going through the usual rebase dance. There’s even a timeline that lets you undo almost anything, which is something I didn’t realize I needed until I had it.
Conflicts are handled differently too. Instead of everything breaking mid-rebase, you can deal with them more calmly, one by one.
It also plugs into GitHub or GitLab, so opening PRs or checking status doesn’t require jumping between tools.
There’s some AI stuff in there as well, like generating commit messages or PR descriptions. Useful sometimes, easy to ignore if you don’t care.
It’s still evolving, so you’ll run into rough edges. And if you’re deeply comfortable with raw Git, you might not feel the need to switch.
But if Git ever feels like messy instead of flow, this makes it a lot smoother.
That’s it
These are the ones I kept using.
You don’t need all of them. Just pick one that fixes something annoying in your workflow and try it.
Top comments (0)