Kilo Code is the community-governed open-source continuation of Roo Code, which was archived on May 15, 2026, with Kilo Code designated as the recommended migration path. Branched in April 2026 and reaching v7.3 by June 12, 2026, it has inherited Roo Code’s 1.5M+ user base and its position as the number one extension by usage on OpenRouter, with over 25 trillion tokens processed. The project is MIT-licensed and governed by its community rather than a single commercial entity.
Key capabilities
500+ model support via BYOK - Kilo Code connects to over 500 models through a bring-your-own-key architecture, supporting Claude, GPT-4o, Gemini, DeepSeek, Llama, Mistral, and any OpenAI-compatible endpoint. This breadth is unmatched among editor-based agentic tools and gives teams complete freedom over model selection and cost control.
Full agentic loop with browser automation - Beyond file editing and terminal execution, Kilo Code includes built-in browser automation, enabling agents to interact with web interfaces, run end-to-end tests, and gather information from the web as part of a coding task — without switching to a separate tool.
MCP and multi-agent support - Kilo Code implements the Model Context Protocol (MCP) for tool and context extension, and v7 introduced parallel agents that can work on separate tasks simultaneously. This makes it one of the few open-source tools that matches the multi-agent capabilities of commercial competitors.
VS Code, JetBrains, and CLI delivery - Kilo Code is available as a VS Code extension, a JetBrains plugin, and a command-line agent, covering the full range of developer environments without requiring a custom editor installation.
Autonomy level
Kilo Code operates at autonomy level 3. Agents can perform multi-file edits, run terminal commands, browse the web, invoke MCP tools, and coordinate across parallel threads, but all changes are surfaced for developer review before being applied. The tool does not auto-commit or deploy autonomously; it is designed for supervised agentic workflows where the developer retains final control.
Strengths
- Active community-governed fork of Roo Code with a clear migration path for existing users
- 500+ models supported via BYOK with no built-in model lock-in
- VS Code, JetBrains, and CLI delivery modes in a single project
- Parallel agents introduced in v7 for concurrent task execution
- MCP support for extending agent capabilities with external tools and context
- Built-in browser automation for end-to-end agentic workflows
- MIT license with full source transparency
- 1.5M+ users and 25T+ tokens processed, providing a large, battle-tested install base
Limitations
- Relatively new as a standalone project, having branched from Roo Code in April 2026 with less than three months of independent history
- Community governance can mean slower prioritization of enterprise features compared to commercially backed tools
- BYOK model means no built-in model access — users must supply and manage their own API keys
- No native Figma or design-to-code integration
- Browser automation is less polished than dedicated browser-agent tools
- Community support rather than commercial SLAs, which may be a barrier for regulated enterprise adoption