Zoo Code: A Roo Code Successor That Pushes Agent Orchestration Further
Zoo Code gives you a whole dev team of AI agents in your code editor.
At a glance
- What is it?
- Zoo Code is a VS Code extension that turns a team of AI agents into an in-editor workflow, forked from Roo Code after that project wound down. It adds semantic code search, safer delegation, and a destructive command guard, but carries the cost of a fast release cycle and a young ecosystem.
- Who is it for?
- Adopt Zoo Code if you are a Roo Code user looking for continued development, or if you want an editor-native agent team with semantic search and safer autonomous runs. Do not adopt it if you need a stable, long-term supported tool, because the project is young, the release cadence is aggressive, and the team is still building migration tooling.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Zoo Code Replaces and Who It Serves
Zoo Code is a VS Code extension that puts multiple AI agents inside the editor, each with a distinct role. It is a direct continuation of Roo Code, a project that the Roo team wound down to focus on Roomote. The Zoo Code team is made up of former Roo contributors, so the core audience is existing Roo Code users who want the same workflow without losing updates. The README positions it as a whole dev team: agents that generate code, refactor, debug, answer questions, and automate repetitive tasks. That framing matters. This is not a chat assistant. It is a tool that tries to manage long-running, multi-step work inside your editor, which is a different category from a single-turn autocomplete or a chat pane.
How the Agent Team Is Organized: Modes and Orchestrator
Zoo Code organizes agents into modes. The built-in modes are Code, Architect, Ask, Debug, and Custom. Each mode has a different scope: Code handles edits and file operations, Architect plans systems and migrations, Ask gives fast answers, Debug traces issues and adds logs. Custom modes let a team build specialized roles for its own workflow. The Orchestrator workflows are the more interesting part. The README describes safer delegation, parallel task coordination, and reliable parent/child task recovery. That means the extension can spawn subtasks and have a parent agent coordinate them, with isolation between subtasks and provider profiles. The key detail is recovery: if a child task fails, the parent can recover it instead of restarting the whole run. That is a real architectural choice, not just a marketing bullet, because it changes how you can trust long autonomous runs.
Semble Codebase Intelligence: Search Without an Indexing Step
One of the headline additions since Roo Code is Semble, described as fast, on-demand semantic code search with automatic setup and no separate indexing workflow. Most semantic search tools require you to build an index before you can query. Semble claims to skip that step, which means you can search your codebase by meaning rather than by string matching, and the setup is automatic. That is a concrete differentiator. The trade-off is that the README does not explain how Semble works under the hood. There is no mention of whether it runs locally, sends code to a remote service, or what the privacy implications are. For an engineering team evaluating this, that missing detail is a reason to ask questions before adopting it for a codebase with sensitive code.
Destructive Command Guard: Autonomy with a Brake
The Destructive Command Guard (DCG) is another addition that targets long autonomous runs. The README says it automatically blocks dangerous commands while trusted work continues without repeated approval prompts. That is a meaningful shift from the usual approve-every-command flow. Instead of pausing on every terminal call, the guard lets the agent keep going unless the command matches a dangerous pattern. The benefit is fewer interruptions and longer stretches of unattended work. The risk is that the guard has to decide what counts as dangerous. If it is too aggressive, it blocks legitimate commands. If it is too permissive, it defeats the purpose. The README does not list what commands are blocked or how to configure the guard, so you will need to check the documentation or test it yourself.
Getting It Running: From Source or VSIX
The README gives clear local setup steps. Clone the repository, install dependencies with pnpm install, then run the extension in development mode by pressing F5 in VS Code. That opens a new window with the extension running, and changes to the webview appear immediately, while core extension changes hot reload. For a production install, you can build a VSIX package with pnpm vsix, which generates a .vsix file in the bin/ directory. There is also an automated command: pnpm install:vsix [-y] [--editor=<command>]. This uninstalls any existing version, builds the latest package, installs it, and prompts you to restart VS Code. You can specify the editor with --editor=cursor or --editor=code-insiders. That is a practical path for teams that want to test a specific commit rather than wait for the marketplace release.
What the Release Cadence Tells You
The repository shows a fast release schedule. v3.80.1 was pushed on 2026-08-29, v3.80.0 on 2026-08-22, and v3.78.0 on 2026-08-15. That is three releases in two weeks, which is aggressive. The v3.80.1 notes list reliability fixes: restored subtask approvals, Vertex Gemini 3.7 empty tool output, terminal startup failures, background service errors, and theme legibility. This is a double-edged signal. On one hand, the team is actively fixing bugs and shipping model updates. On the other hand, a cadence this fast means you are always chasing changes. For a tool that runs autonomous agents in your editor, stability matters. You do not want a background service error to silently break a long-running task. The release notes suggest that kind of issue is still being ironed out.
The Migration Path from Roo Code and the Alternative
The README links to a Roo to Zoo migration guide at docs.zoocode.dev/roo-to-zoo-migration. The team explicitly says they plan to help users transition, and they point to Reddit and Discord for support. That is a real migration path, but it is also a sign that Zoo Code is not a drop-in replacement. You will need to change your workflow, even if slightly. The alternative to Zoo Code is to stay on Roo Code or move to Roomote, the project the original Roo team is now focused on. Roomote is not an editor extension in the same way; it is a separate product. The difference in approach is that Zoo Code keeps the agent team inside VS Code, while Roomote appears to be a different environment. If you value the editor-native workflow, Zoo Code is the closer match. If you want the team that built the original to keep evolving the concept, Roomote is the direction they chose.
Licence, Maintenance Cost, and What to Verify First
Zoo Code is licensed under Apache-2.0, which is permissive for commercial use and modification. That is a plus for companies that want to fork or embed parts of the code. The maintenance cost is tied to the release cadence: you will need to update frequently to get bug fixes, and you will need to track changes to the configuration and modes as the project evolves. The README mentions new providers including Zoo Gateway, Moonshot, Kimi Code, Kenari, Friendli, and OpenCode Go, plus support for Claude, GPT, Gemini, Kimi, GLM, Grok, and MiniMax. That breadth is good, but it also means the provider integration surface is large, and each one can break. Before you adopt, verify that your specific model provider is on the supported list and that the version you install works with your editor. Also check the migration guide if you are coming from Roo Code, because the README does not promise a seamless transition.
Editorial conclusion
Adopt Zoo Code if you are a Roo Code user looking for continued development, or if you want an editor-native agent team with semantic search and safer autonomous runs. Do not adopt it if you need a stable, long-term supported tool, because the project is young, the release cadence is aggressive, and the team is still building migration tooling. Before adopting, verify that your preferred model provider is supported, test the migration guide from Roo Code, and check whether the Destructive Command Guard blocks commands you actually need for your build pipeline.
Community notes