A tool I had been using daily for months was shut down in May 2026. Not because it was bad — but because its team believes the IDE era is over. Three million downloads, 24,000 GitHub stars, irrelevant. That was Roo Code. What the community rescued from the ashes is Zoo Code.
First: What Is This Thing?
In short: an AI that sits right inside your programming editor and turns your instructions into code. But under the hood there's something more interesting.
At its core, such an assistant is a collection of really well-crafted system prompts — built-in instructions to the language model that you, the user, never get to see. These prompts tell the model how to think, plan and respond. Your input ("build me a login form") gets combined with those templates, the model produces code from it — and the assistant is also allowed to read files, search the project, and run commands.
That's the whole magic: clever templates plus your wish plus access to your files. No sorcery, but seriously useful.
Roo Code, Zoo Code — What Happened Here?
Roo Code was exactly such an assistant: open source, for VS Code, you bring your own model. Three million downloads, 24,000 GitHub stars. A real community around it — and a tool I had used heavily for a long time.
Then, in April 2026, the announcement: the team was shutting down the plugin. Not for lack of success — but because they believe the IDE era is over. They're betting everything on Roomote, a cloud agent that lives in Slack, GitHub and Linear and carries out tasks end-to-end on its own.
Whether they're right, time will tell. For everyone who still wants a solid assistant in their editor, Zoo Code was born: a community fork that picked up where Roo Code left off on May 16, 2026 — one day after the shutdown. Same features, same settings, same licence (Apache 2.0).
Zoo Code is available as a plugin for VS Code and its open-source offshoots like VS Codium. You install it, give it access to an AI model, and from then on the AI sits in your editor.
Not Just Code
A misconception up front: these tools are built for programming, but they can do far more. Anything that's a structured text document is home turf for them: reports, analyses, form letters, tables, documentation.
The trick: the assistant sees the files in your working folder. So it doesn't answer abstractly — it works with your real data.
An example — and the honest privacy note that goes with it. Imagine you put your last twelve months of bank statements into a folder and ask:
Read all the bank statements in this folder. Build an overview of my spending by category, find recurring subscriptions and items I've probably forgotten, and tell me the three biggest levers to save money each month.
The result can be impressive — a financial analysis that would otherwise take you hours, or an advisor.
But: bank statements are among the most sensitive data you own. Whether you send a prompt like that to a cloud AI that reads along is something to think hard about. And this is exactly where Zoo Code's next property becomes worth its weight in gold.
The Upsides: Control
Zoo Code is open source — the code is out in the open, anyone can check what happens. And above all: you decide which model does the work.
- Full control: you bring your own model access — via OpenRouter, directly with a provider, or through a local model.
- Local with Ollama — in theory: with Ollama the model runs on your own machine; your data never leaves it. Honestly, though: on typical laptop hardware this is too slow for serious work. On a proper server with its own graphics card it's a different story — then the financial analysis above suddenly becomes safely doable, and with it sensitive domains too, from the tax office to critical infrastructure.
- Codebase index with Qdrant: Zoo Code can make your whole project searchable — via a vector database (a database that searches by meaning, not just exact words) called Qdrant, which can likewise run locally. That way the AI finds the right context even in huge codebases.
The Downsides: Convenience, Cost — and an Honest Warning
This freedom has a price — literally and figuratively.
- No ecosystem: while providers like Anthropic build a whole universe around Claude (assistant in the editor, Cowork, scheduled routines and more), Zoo Code is "just" the assistant in the editor. Powerful, but on its own.
- No subscription: there's no flat monthly fee that covers everything. Instead…
- Token costs can explode: you pay per token, i.e. per processed unit of text. With heavy use it adds up — and can really hurt at month's end (more on that in the Copilot article).
- API key required: you need an API key — an access code to a model provider (OpenRouter, or many others). A bit of setup is part of the deal — not double-click-and-done.
- Open source is no guarantee of permanence. Roo Code proves this clearly: three million downloads, 24,000 stars — and still shut down. The maintainers moved on. A community fork can soften the blow, but whether Zoo Code will still be actively maintained in two years is an open question. Anyone building long-term on a tool should factor that in.
How I Use It Myself
I had been using Roo Code for a while — my reliable fallback when Claude Code got stuck. When Roo Code was shut down in May 2026, Zoo Code appeared on the Marketplace the very next day. For me, the migration was trivial: export settings from Roo Code, install Zoo Code, import — three mouse clicks. Chat history was gone, but everything else carried over: API keys, model configurations, all set up and ready to go.
My most common use today: when Claude Code goes nowhere or runs in circles, I just switch to the second plugin in the same editor: Zoo Code. I carry on at exactly the same spot, no subscription, purely usage-based. For me that usually lands between €5 and €20 a month. Once Zoo Code has untangled the knot, I switch back to Claude.
Two assistants side by side, each with its strengths — that's exactly where the open architecture is worth its weight in gold.
Who Is It For?
Zoo Code is not for your very first step. But if you want to understand what happens under the hood, need full control over your data, or already work with your own servers, it's an excellent tool.
For everyone else it's above all one thing: a good example that "AI in the editor" doesn't have to mean "a US corporation reads along" — and that an active community is sometimes the only guarantee open source needs.
Unpaid recommendation, my own opinion — as always here.