Tabby

If you work at a company with strict rules about where source code can go, most AI coding assistants are a non-starter. GitHub Copilot and its peers send your code to someone else's servers, and that's often enough for a security team to block the whole category. You end up choosing between giving developers AI help and keeping your codebase private. Tabby exists to remove that choice. It's an open-source, self-hosted coding assistant you run on your own hardware, and the project's pitch is blunt: "no need for external DBMS or cloud services."

The core features will look familiar if you've used any modern assistant: code completion that "predicts your next move," an Answer Engine for coding questions inside your IDE, inline chat, and data connectors that pull in context from your own sources. The demo on tabbyml.com shows the Answer Engine reading 12 files in the TabbyML/tabby repo to confirm it uses Tailwind CSS, then citing the specific files. That's the kind of repo-aware answer you want from a tool that lives next to your code. Tabby also says it runs on consumer-grade GPUs and works with Cloud IDEs, so you won't need a data center to try it.

The bigger point is control. Because the code is open (the GitHub repo sits at 21.6K stars), your security team can audit it, and Tabby frames this as software supply chain safety. The free Community tier covers up to 5 users with local deployment and the full core feature set, which is generous for a small team. Larger teams move to the $19 per seat Team plan or a custom Enterprise deal. There's also a newer agent called Pochi, billed as a "full-stack AI teammate" that plans, executes and checks in on tasks, though the site says little about how it works. If your main problem is getting AI help without your code leaving the building, Tabby is one of the more credible answers. Just know that self-hosting means you own the setup, the hardware and the upkeep.