Twinny

Software engineer Roderick MacArthy created Twinny after finding that existing coding assistants required third-party logins, hosted proxies, or complicated setups for local models. He built the extension to run directly against local Ollama instances and custom endpoints without telemetry or middleman servers. That origin keeps Twinny focused on a clear goal: free, offline-capable code completion and chat inside standard Visual Studio Code.

Why Twinny instead of Cursor AI

Cursor AI charges twenty dollars per seat monthly alongside metered fees for high-volume model usage. For engineering teams managing infrastructure across repositories, mandatory per-seat SaaS bills create continuous budget overhead. Cloud assistants also route proprietary code through external infrastructure, raising compliance issues unless enterprise contracts are in place. Twinny eliminates subscription expenses by decoupling the editor from model inference. The extension carries no licensing fees and runs inside stock Visual Studio Code or VSCodium, connecting directly to local Ollama runtimes or private servers. Running models like Codestral on local GPUs incurs zero token fees and zero subscription overhead.

Cursor holds an edge in autonomous multi-file refactoring and project-wide agent workflows. Because Cursor is an editor fork, it embeds repository indexing and automated multi-file edits directly into the editing loop. Twinny stays within standard editor boundaries, focusing on inline Fill-in-the-Middle code completion, side-by-side diff previews, and conversational chat. Teams that do not need autonomous repository refactoring avoid seat fees while keeping their code on local hardware.

What it does well

• Direct local inference via Ollama or custom endpoints, keeping infrastructure templates and application logic inside private network perimeters.
• Fill-in-the-Middle code completion that predicts code between existing blocks, speeding up boilerplate for Kubernetes manifests and Terraform definitions.
• Side-by-side diff views that let developers inspect and selectively accept modifications before committing changes to files.
• Workspace document embeddings that provide contextual awareness across project files without exporting codebases to cloud indexing services.
• Customizable prompt templates for recurring tasks such as drafting unit tests, adding TypeScript types, and writing Git commit messages.