AI coding
Twinny
Twinny is a self-hosted AI coding assistant designed specifically for VS Code.
What is Twinny?
Twinny is a self-hosted AI coding assistant designed specifically for VS Code.It enables developers to run their own models on local laptops, desktops, or a team gateway, supporting Ollama, llama.cpp, LM Studio, or any OpenAI-compatible server.Its core philosophy is that "nothing leaves your machine unless you point it somewhere," ensuring code privacy and security.
The tool offers features such as code autocomplete, chat, inline editing, code review, and agent mode.All requests are sent only to the server you configure, with no telemetry collection and no account required.Twinny is open source under the MIT license, making it suitable for individual developers seeking maximum privacy as well as teams needing centralized model access management via a unified gateway.
Key features
- Local AutocompleProvides fill-in-the-middle ghost text completion as you type, working well even on smaller models.
- Contextual ChatAttach files, symbols, the problems panel, git diffs, terminal output, or the workspace index using @ mentions.
- Inline EditTrigger with Ctrl+I to describe changes, accepting or rejecting each hunk individually; nothing is kept until confirmed.
- Smart Code RevieReview the working tree, a branch against its base, or GitHub pull requests to assist in identifying potential issues.
- Agent ModeReads, searches, edits files, and runs commands with each step visible; commands ask for permission first, and file changes can be undone.
- Terminal IntegraGenerates terminal commands from descriptions, showing them before execution, with failures traced to specific files and lines.
- Workspace IndexMaintains a keyword and vector index of the workspace, reranked and updated on save to improve code retrieval relevance.
- Zero Telemetry &No accounts, no telemetry data upload, source code is open under the MIT license, and the gateway never calls home.
How to use it
- Install the Twinny extension from the VS Code marketplace or via VSX.
- Set up a model backend on your local machine, such as installing Ollama, llama.cpp, or LM Studio.
- Configure the model server address in Twinny settings, ensuring requests point to your local instance or designated gateway.
- Start coding to experience autocomplete; use Ctrl+I for inline edits, or reference context in chat using @ mentions.
- (Optional) For teams, deploy the twinny-server gateway to centrally manage keys and usage.
Where it fits
- Privacy-SensitivIdeal for regulated industries like finance or healthcare, ensuring proprietary code and data remain entirely on internal networks or local devices.
- Offline Coding EProvides continuous code assistance and completion in environments without stable internet connectivity by relying on locally running models.
- Team CollaboratiOffers model access to the entire development team through a single gateway, centralizing API key management, policies, and usage statistics.
Pricing and free quota
Free for up to 5 developers forever; Personal one-time fee of $29; Team plan at $15 per seat/month (billed yearly) for seats beyond the free 5; Enterprise at $25 per seat/month from 50 seats.
Who it is for
- Software engineers prioritizing data privacy
- Development teams requiring localized AI assistance
- Organizations restricted by compliance requirements
- Power users wanting custom model backends
FAQ
Does Twinny collect my code data?
No.Twinny has no telemetry.Requests go only to the server you set and nowhere else unless you point it somewhere.
Which model backends are supported?
It supports Ollama, llama.cpp, LM Studio, and any OpenAI-compatible server.
How can teams share model resources?
Teams can deploy the twinny-server gateway to pool teammates' computers as a backend or manage model access and keys centrally.
What are the limits of the free version?
The free version supports up to 5 seats forever, including the full extension and gateway functionality.