00:00:00If you're using Copilot right now, your code might already be training someone else's model.
00:00:04You install Copilot, it works great, and you move on, but parts of your codebase can actually
00:00:09leave your machine.
00:00:10That can be a problem.
00:00:12This is Tabby.
00:00:13An open source alternative to this that gives us the highest level of privacy compared to
00:00:17things like Copilot, Tab9, and Cursor.
00:00:20We can get the same speed, same autocomplete, same workflow, and our code never leaves our
00:00:25machine.
00:00:26That's basically Tabby.
00:00:27I'll show you how to set it up and how to get it working in the next few minutes.
00:00:36Now at a simple level, Tabby is a self-hosted AI coding server.
00:00:40You run it locally, usually with Docker, you pick the model you want, and then you connect
00:00:44it to your IDE.
00:00:45That's it.
00:00:46You get real-time code completions in Codebase Aware Chat, just like you'd expect.
00:00:50But the real reason devs care isn't just the features, it's the control we get.
00:00:55Your code stays inside your network without any subscriptions, and it works fully offline.
00:01:01It's built for teams with things like SSO, RBAC, and audit logs, and it's been blowing
00:01:05up on GitHub with over 33,000 stars for a good reason.
00:01:09Honestly, though, none of these matter if it feels bad to you, so let's skip all this stuff
00:01:13and just jump straight into the demo.
00:01:15If you enjoy these types of tools to speed up your workflow, be sure to subscribe to the
00:01:19channel.
00:01:20We have videos coming out all the time.
00:01:22Here's what the setup actually looks like.
00:01:24You run one Docker command, and Tabby is up and running locally.
00:01:28Then you install the VS Code extension, point it to your local server, and you're done.
00:01:34Now you're getting multi-line completions right inside your repo.
00:01:38So here in Tabby, I can open up to check out the models that I'm using, and you can see
00:01:42here that these are the three that we're using, and they're running locally.
00:01:45No clod or open AI where your data is going.
00:01:48Over in VS Code, I can start with a rough function, and just with the tab button, Tabby will autocomplete
00:01:53this for me.
00:01:55I can push it a bit by chatting with it on the side to optimize and expand on my current
00:02:00code as well.
00:02:01It's all pretty simple and straightforward.
00:02:03I can highlight some code and ask it to refactor performance or add tests.
00:02:07It responds instantly, and it understands your repo context, not just a single file.
00:02:12I can even drop in a comment of something I want built, and you can see it picks up right
00:02:16here and actually builds it out for me.
00:02:19Now over in local host, Tabby is still connected to everything in VS Code, so I can read my
00:02:23code chats, even expand on it and follow up and ask questions.
00:02:27This is all saved right here in local host.
00:02:30No cloud, no data leaving your machine, and it feels very similar to Copilot, except, now
00:02:35this is a big except, we actually own everything.
00:02:37All right.
00:02:38I kept the demo quick because honestly, it was just that simple to fire up and get going.
00:02:43Now let's talk about why this actually matters in our real world workflow.
00:02:47Now the real issue with cloud AI tools isn't that they're bad.
00:02:51It's that the trade-off that we get is hidden, right?
00:02:53With cloud tools, your code may be used to train their models.
00:02:57With Tabby, your code never leaves your own network, right?
00:03:01Cloud tools, you're paying per developer every month because it's free forever.
00:03:05Well, it's not, right?
00:03:07We're paying for it.
00:03:08That's what we get.
00:03:09And with cloud tools, we also need the internet.
00:03:11With Tabby, I'm not paying for it, it's running offline, and this shows up in real work.
00:03:16So really we get less boilerplate, we can refactor messy legacy code with less hesitation.
00:03:22We can learn frameworks quicker, generate tests and docs without jumping between all these
00:03:26tools.
00:03:27So really this is less wasted time, hopefully less risk, and a lot more control over how
00:03:33we work.
00:03:34That's why a lot of privacy-focused devs or teams are starting to move away from these
00:03:38cloud-first tools into tools like this.
00:03:41Now let's compare it to other options because that's really what you guys want to hear, right?
00:03:45Tabby is the easiest.
00:03:47It's great quality, almost no setup, but it does live in the cloud.
00:03:50We have continued dev.
00:03:52It's flexible, it's local-first, but it's more of a power user tool.
00:03:56Tab 9 is more enterprise-focused, and then obviously now I'm here talking about Tabby,
00:04:01which is self-hosted, it's free, a lot higher privacy, and it is built for teams.
00:04:05But the real difference is this, Tabby is not just a plugin, it's a dedicated AI coding server.
00:04:11That really changes everything.
00:04:12You get a co-pilot-like experience, the flexibility people like in Continue, and team-level controls
00:04:19that other users usually charge for.
00:04:21So instead of renting access to AI, we actually own the infrastructure behind it.
00:04:26Now let's be honest, right?
00:04:28Because people love a lot of things, but it's open source, is that enough to actually make
00:04:32the switch?
00:04:33Well, the setup is pretty quick, usually just a Docker spin-up, and then it fades into your
00:04:39workflow.
00:04:40When you get locked into a single model, you can choose the model, and overall it feels
00:04:44much more production-ready now than it did before.
00:04:47Now again, open source, there's downsides.
00:04:50The quality depends on the model you choose, so smaller models aren't going to be as powerful,
00:04:55and hardware does matter.
00:04:56If you want a smooth performance, a GPU is going to help a lot.
00:04:59I'm running all this on a Mac M4 Pro, and it felt pretty good.
00:05:04The setup is still more work than cloud tools, so it's not ideal for anyone who's non-technical,
00:05:09but you're watching this.
00:05:10Assuming you are.
00:05:11And of course, like any AI tools, you still need to review the code.
00:05:14This leads me to the question that we actually want answered.
00:05:17Is this worth using?
00:05:19Yes, kind of, but it depends on a few things.
00:05:22You should use Tabi if you care about privacy, you hate subscriptions, you work in a regulated
00:05:27environment, or you need something your whole team can rely on.
00:05:30In those cases, this is an awesome choice to try to integrate into that workflow, but if
00:05:35you want the absolute best model with zero setup, no effort, honestly, come on, cloud
00:05:40tools are still easier.
00:05:41The difference now is the trade-off has changed.
00:05:43We're not choosing between a smart cloud tool and a weaker local one anymore, you're choosing
00:05:48between convenience with something like cursor, or strong enough AI on your own terms.
00:05:54And for a lot of developers, this is starting to matter more and more.
00:05:58Tabi isn't trying to be the smartest AI.
00:06:01It's trying to be the one we can actually maybe trust.
00:06:04I've linked some docs and repos in the description.
00:06:06If you enjoy open source and other AI tools like this one, be sure to subscribe to the
00:06:11Better Stack channel.
00:06:12We'll see you in another video.