twinny
twinny is a free, MIT-licensed Visual Studio Code extension for AI code completion and chat. It defaults to a local OpenAI or Ollama-compatible endpoint, so completions run on your own machine and it works offline, and it also supports OpenAI, Anthropic, OpenRouter, DeepSeek, Cohere, Mistral, Perplexity and Groq. It does fill-in-the-middle completions, sidebar chat, side-by-side diffs and commit messages, and it is funded by donations.
What does twinny do?
twinny is "a free AI extension for Visual Studio Code, offering powerful AI-assisted coding features." Its centre is completion: "fill in the middle code completion", single and multiline, shown as real-time suggestions as you type. The sidebar chat lets you "discuss your code with AI, getting explanations, tests, refactoring suggestions, and more." Around those sit direct acceptance of code solutions, a side-by-side diff view, creating new documents from code blocks, git commit message generation, workspace embeddings for context, and customisable prompt templates.
The default provider is a "localhost OpenAI/Ollama Compatible API", which is what makes it an offline tool: point it at a local model server and nothing leaves the machine. Eight hosted providers are also listed: OpenAI, Anthropic, OpenRouter, DeepSeek, Cohere, Mistral AI, Perplexity and Groq. The README states "online and offline operation" as a feature. There is no agent mode described, no command execution, and no MCP.
Key facts
- Licence: MIT, copyright Richard Macarthy, with no added conditions.
- Editor: Visual Studio Code only, installed from the marketplace.
- Fill-in-the-middle completion, single and multiline, plus sidebar chat, diff view, document creation from code blocks and commit messages.
- Default provider: a localhost OpenAI or Ollama-compatible API; hosted options are OpenAI, Anthropic, OpenRouter, DeepSeek, Cohere, Mistral AI, Perplexity and Groq.
- Online and offline operation, and workspace embeddings for context-aware suggestions.
- No agent mode, command execution or MCP is described in the README.
- Free with no paid tier; the README asks for donations.
- Latest tagged release when read: v4.0.3 on 2026-09-08; 3,648 stars on 2026-09-18.
What does it replace, and where does it fall short?
twinny replaces GitHub Copilot's core, inline completion and chat inside VS Code, with a model you run yourself and no subscription, which is why it sits on the open-source alternatives to GitHub Copilot page as the offline pick. Against Cursor it covers completion and chat but none of the agent features.
Where it falls short: VS Code only, no agent that edits files or runs commands, no MCP, and the quality of completion depends on the local model you choose and the hardware it runs on. The README calls the project "actively developed and provided as is" and says functionality may vary between updates. For an agent as well as completion, Kilo Code or Cline.
How does twinny run?
Install from the VS Code marketplace. By default it talks to a local OpenAI or Ollama-compatible server, so run Ollama or a similar server with a code model and it works with no account. For the hosted providers you add your own key. There is no hosted twinny and no paid tier; the README lists a Bitcoin address for donations.
Who is twinny for?
A VS Code user who wants Copilot-style completion from a local model, offline, for free. Anyone who wants an agent to make multi-file changes should pick a different entry on the same page.
What limits does the README state?
From the README: "Twinny is actively developed and provided as is. Functionality may vary between updates." Known issues are tracked on the GitHub issues page rather than listed.
Questions people ask
Is twinny open source?
Yes. twinny is MIT licensed, an OSI approved licence with no use restrictions, and its README says it is free and open source. Development is funded by donations rather than a paid tier.
Does twinny work offline?
Yes. The default provider is a localhost OpenAI or Ollama-compatible endpoint, so with a local model server running the completions and chat never leave your machine. The README lists online and offline operation as a feature; the hosted providers are optional.
Which models can twinny use?
Any model served through a local OpenAI or Ollama-compatible API, which is the default, or the hosted providers it lists: OpenAI, Anthropic, OpenRouter, DeepSeek, Cohere, Mistral AI, Perplexity and Groq. The README does not name specific models.
How does twinny compare with GitHub Copilot?
twinny does the inline fill-in-the-middle completion and the chat that Copilot is known for, in VS Code, from a model you run for free. Copilot adds an agent, a hosted model and coverage across GitHub. The open-source alternatives to GitHub Copilot page compares it with Tabby, Cline and Kilo Code.
Does twinny edit files or run commands like an agent?
Not as described in its README. It accepts code solutions directly, shows a side-by-side diff, and creates new documents from code blocks, but there is no autonomous agent mode, no terminal execution and no MCP support listed. It is a completion and chat tool.
Sources
- twinny README and MIT LICENSE: github.com/twinnydotdev/twinny, read 2026-09-17.
Compared with the others
On the open-source alternatives to GitHub Copilot page, twinny is the pick for ghost text in vs code from a local model. MIT, fill-in-the-middle completion and chat, defaulting to a local Ollama-compatible endpoint, offline, free.
Also on that page: Tabby for a self-hosted copilot server for a team, Kilo Code for completion and an agent in one extension, Cline for the chat and agent side, any model.
On the open-source alternatives to Cursor page, twinny is the pick for free completion from a local model. MIT VS Code extension with fill-in-the-middle completion and chat, defaulting to a local Ollama-compatible endpoint and working offline.
Also on that page: Zed for replacing the editor itself, Kilo Code for autocomplete plus agent, inside vs code or jetbrains, Cline for an agent with reviewable diffs and any model.
On the open-source alternatives to Windsurf page, twinny is the pick for free local completion in vs code. MIT, fill-in-the-middle completion from a local Ollama-compatible endpoint, offline.
Also on that page: Zed for an open editor that hosts other agents, Kilo Code for autocomplete plus agent in vs code or jetbrains, Cline for an agent with reviewable diffs.
More coding assistants and agents
An MIT terminal coding agent with a full-access build agent and a read-only plan agent switched with the Tab key, installers for every platform, a beta desktop app, and the most-starred repository in this directory.
OpenAI's Apache-2.0 coding agent that runs locally in your terminal, with a desktop app and IDE extensions, signed in with a ChatGPT plan or an OpenAI API key, so the code is open and the models are OpenAI's.
Google's Apache-2.0 terminal agent for Gemini models with a free tier of 1,000 requests a day on a personal Google account, built-in search grounding, file and shell tools, MCP support, checkpointing and a GitHub Action.
A high-performance code editor from the creators of Atom and Tree-sitter, GPL-3.0-or-later with Apache-2.0 components, with an agent panel, external agents over ACP, edit prediction, MCP tools, sandboxing, and local models through Ollama or llama.cpp.
An MIT self-hosted control centre, called Agent Canvas, that runs the OpenHands agent or Claude Code, Codex, Gemini and any ACP-compatible agent across local, Docker and cloud backends, with automations triggered by schedules and webhooks.
An Apache-2.0 coding agent for VS Code, JetBrains, the terminal and a desktop app that edits files as reviewable diffs, runs commands with your approval, works with any model provider including Ollama, and connects tools through MCP.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/twinnydotdev/twinny.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.