A blog for a coding assistant that never phones home
What twinny is, what this blog is for, and the rule every post here follows.
twinny is an open-source coding assistant for VS Code. Autocomplete, chat, inline edit and code review, against a model server you run: on your laptop, on a desktop with a GPU, in your data centre, or pooled from your teammates’ computers. It has been in development in public since 2023, is MIT licensed, and has been installed a couple of hundred thousand times across the Marketplace and Open VSX.
The unusual part is what it does not do. It makes no request to anything except the server you configured. No account, no telemetry, no update check. The team gateway, twinny-server, checks its licence with a public key it ships with and never contacts us. That constraint shapes almost every design decision in the project, and those decisions are what this blog is for.
What to expect
Short posts, from the code, about how things work:
- How a completion is built, and why a 7B model is enough for it.
- How a team shares one GPU without opening a port on anyone’s machine.
- Why the licence check has no server, and what happens when a licence lapses.
- Using the desktop’s GPU from the laptop over a peer-to-peer link, with no relay.
- What each release changed and why, at more length than a changelog line.
The rule
Everything written here matches the documentation. If a post says the gateway does something, the gateway does it, today, in the version you can install. No roadmap posts, no benchmarks against other tools, no “the best model for X”. Where a model or a runtime is named, it is because it works and the docs say how.
If you want the product rather than the notes, it is at twinny.dev. If you want to argue with a post, discussions is the place.