The on-premises coding assistant that skips the cloud bill

What it does Tabby is a Rust-based AI coding assistant that serves inline completions and chat through an OpenAPI interface. It runs entirely on-premises—no external database or cloud service required—and targets consumer-grade GPUs so a single machine can handle team workloads. IDE plugins wire it into VS Code, Vim, and JetBrains, while a web-based Answer Engine indexes internal repositories to answer engineering questions.

The interesting bit Instead of treating self-hosting as a downgrade from SaaS, Tabby packages enterprise controls that closed-source assistants often lock behind cloud contracts: LDAP authentication, team management, usage analytics, and RAG-based completions that pull context from your actual codebase, local LSP declarations, and recently modified files.

Key highlights

• Fully offline operation with no DBMS or external cloud dependency
• Consumer-grade GPU support for both CUDA and Apple Metal
• OpenAPI interface for custom integrations and Cloud IDEs
• RAG-based completion using repository context, local LSP symbols, and recent edits
• Enterprise features including LDAP auth, GitHub/GitLab SSO, and team usage reports

Caveats

• Advanced agentic features (such as automated GitHub issue resolution) require a separate product, Pochi, or a private-preview waitlist
• The README is structured as a chronological changelog, so discovering current capabilities requires some scrolling archaeology

Verdict Teams with compliance or privacy constraints that prevent sending code to third-party APIs should look here; individual developers already happy with Copilot’s cloud convenience likely won’t want the infrastructure overhead.