What is twinny?

This open-source VS Code extension adds AI-powered coding features that run entirely on your own hardware. It offers ghost-text completions, inline diff-based edits, a chat sidebar that can reference files and symbols, and an optional agent mode that can modify files and run commands. A built-in workspace index combines keyword and vector search for context. It also handles code review, commit-message generation, terminal command suggestions, and user-editable prompt templates. The extension talks to any OpenAI-compatible server you run — Ollama, LM Studio, llama.cpp, or a hosted API — and can also connect to a self-hosted team gateway (twinny-server) that shares models across developers, enforces policy, and adds PR-review plugins, SSO, shared indexes, backups, and audit logs. The gateway is deployed via npx, Docker, or Helm. Free for solo use and teams up to five seats; larger teams buy licences. No telemetry, no required accounts, MIT licence.

Should you self-host twinny?

twinny scores 63/100 on selfhost.directory (how scoring works) and ranks #16 of 120 in AI Assistant. It has 3 684 GitHub stars.

It shipped 1 stable release in the past year. The latest, v4.0.20, came out on 22 September 2026.

Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is MIT-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is hermes-agent (89/100, 257 926 stars). It scores higher overall, so compare the two before you pick. See twinny vs hermes-agent side by side.

Features

Agent modeExperimental
Audit Log
Chat
Code Completion
Code Review
Commit messages
Inline edit
OIDC SSORequires paid team licence
P2P device sharing
Prompt templates
PR review pluginsRequires paid team licence
Shared context indexRequires paid team licence
Team gateway
Terminal
Workspace index

FAQ

Does twinny run an agent that edits my files on its own?

Only if you switch on agent mode, which is off by default. Then the chat model edits files and runs commands to finish what you asked.

Can I share one GPU machine with my laptop?

Yes, that is what Devices is for. No port forwarding, no account.

Does it work over Remote SSH, WSL or Dev Containers?

Yes, twinny runs on the remote side so localhost means the remote machine.

Can I use it with Cursor, Windsurf or VSCodium?

VSCodium and other Open VSX builds work. Forks of VS Code that support standard extensions generally work but are not tested.

Which model should I use?

For a quick answer: qwen2.5-coder:1.5b-base for completion and qwen2.5-coder:7b-instruct for chat on Ollama.

Does it work offline?

Yes, with a local server. Everything including the workspace index and its reranker runs on your machine.

Does my code leave my machine?

Only to the server you configured. With a local server, nothing leaves. With a paired device, prompts go to your other machine over an encrypted link. With a hosted API, prompts go to that vendor.