Kilo Code

Kilo Code is an open-source AI coding agent available through VS Code-compatible editors, JetBrains IDEs, a command-line interface, and managed cloud-agent workflows. In an editor, it can inspect repository context, propose a plan, edit multiple files, run commands, and present changes for review. The canonical repository is published under the MIT License, so teams can inspect, modify, and redistribute the extension under those terms.

Kilo supports multiple model providers rather than requiring one model family. Individual platform access is free and open source. Model inference is separate: users can choose free or local models, bring provider credentials, or use Kilo-managed inference. Cloud compute is billed separately, while Teams and Enterprise add organization, governance, security, and support features.

Top features

• Editor agent: reads project files, writes code, applies multi-file changes, and runs development commands from a reviewable IDE workflow.
• Named agents: Code, Plan, Ask, Debug, and Review provide task-specific workflows, with custom agents available for additional instructions.
• Provider choice: supports Kilo-managed access, bring-your-own-key providers, local models, and compatible model endpoints.
• CLI and cloud options: terminal use supports shell-centered development, while cloud agents can take bounded work outside a continuously open local editor.
• MCP connectivity: Model Context Protocol servers can add external tools and project context beyond the local repository.
• Open-source licensing: the MIT repository provides a public implementation and inspectable extension codebase.
• Review loop: proposed edits and command results remain visible, preserving normal diff, test, and git practices.

Use cases

• Plan and implement a feature that touches application code, configuration, tests, and documentation.
• Investigate a failing build by inspecting logs, tracing relevant files, applying a patch, and rerunning checks.
• Use a coding agent in VS Code, a compatible editor, JetBrains, or a terminal without standardizing on one model vendor.
• Delegate a contained backlog item to a cloud agent, then review the resulting branch or pull request.
• Connect issue trackers, documentation, or internal utilities through an approved MCP server.