Roo Code has shut down. Your models still need an editor.
Roo Code shut down on 15 May 2026 and its repository is archived. If you used it with a model server of your own, twinny runs against the same servers, has an agent that edits files and runs commands, and adds autocomplete as you type and a gateway for teams.
- 01
Install twinny from the Marketplace or Open VSX. With a local server running, it finds it at first start and sets up chat, autocomplete and embeddings from the models you have.
- 02
Add any other server you had in Roo as a provider: same host, port, model name and key. Hosted APIs are in the presets.
- 03
Turn on agent mode with Shift+Tab in the chat. Choose whether edits apply at once or open as a diff, and whether commands ask first.
An agent with a leash
It lists, reads and searches files, asks the language servers, edits, renames and runs commands. Edits can open as a diff to accept; commands ask by default, or run from an allow list you build.
Autocomplete as well
A small fill-in-the-middle model suggests code as you type, beside the chat model doing the agent work.
Runs on what you run
Ollama, LM Studio, llama.cpp, LiteLLM or any OpenAI-compatible server. Agent mode wants a capable coder model with 8k tokens of context or more.
One gateway for the team
twinny-server sits on the machine with the models and gives every developer a key, usage per person, policy and an admin page. Free for 5 developers, $15 a seat after that, 30-day trial.
- Does my Roo Code install stop working?
- The extension was shut down and its repository archived, so it gets no more fixes, including for new versions of VS Code and of the model servers.
- Does twinny have modes like Roo’s?
- Two: chat, and agent mode, a switch on the chat (Shift+Tab). There are no custom modes with their own prompt and tools; the system prompt is one template you can edit.
- Does twinny support MCP?
- Yes, in agent mode. Name your MCP servers in the twinny.chatToolsMcpServers setting and their tools join the agent’s own, each call asking before it runs. Servers run on your machine over standard input and output; HTTP servers are not supported yet.
- Is agent mode finished?
- It is marked experimental in the docs. Small models call tools badly; use a capable coder model.
Everything else is on the home page: features, the team gateway, security and pricing.