Kilo Code Brings Open-Source Agentic Engineering to Your IDE

AI-powered coding assistants have quickly evolved from simple single-file autocomplete widgets into complex, multi-file agentic systems. However, many of the leading tools in this space operate as closed-source black boxes, locking developers into specific models, proprietary platforms, and marked-up API pricing.

Kilo Code is taking a different approach. As an open-source agentic engineering platform, Kilo Code aims to give developers complete control over their AI stack. By supporting over 500 models, offering zero-markup pricing, and integrating directly into existing workflows across the CLI, editors, and CI/CD pipelines, Kilo provides a highly configurable alternative to proprietary coding assistants.

Meet the Specialized Agents

Rather than relying on a single general-purpose prompt to handle every task, Kilo structures its platform around specialized agents. Developers can switch between these agents depending on the specific phase of their development workflow:

• Code: The default agent. It focuses on implementing and editing code across multiple files using natural language instructions.
• Plan: Designed for the architectural phase. This agent drafts implementation plans and designs system architecture before any actual code is written, helping prevent the AI from jumping straight into incorrect implementations.
• Ask: A read-only agent that answers questions about the codebase. It analyzes the repository structure and logic without modifying any files.
• Debug: Built specifically to troubleshoot, trace, and resolve active bugs or errors in the application.
• Review: An automated reviewer that scans code changes to surface potential issues across performance, security, style, and test coverage.

Beyond these built-in agents, the platform is designed to be extensible, allowing teams to build and configure their own custom agents tailored to internal libraries, coding standards, or proprietary APIs.

Multi-Model Flexibility and Open Pricing

One of Kilo's most compelling features is its "open pricing" model. Instead of paying a flat, marked-up subscription fee for a restricted set of models, developers can choose from over 500 models—including cutting-edge options like GPT-5.5, Claude Opus 4.7, Claude Sonnet 4.6, and Gemini 3.1 Pro Preview.

Kilo allows developers to switch models mid-task, making it possible to use a fast, low-cost model for initial drafting or basic autocomplete, and then swap in a high-reasoning model for complex debugging or architectural planning. Users pay the model provider's direct rate with zero markup, and no API keys are required to get started.

Editor Integrations and CLI Setup

Kilo is designed to meet developers wherever they prefer to work, offering native integrations across major IDEs and terminal environments.

VS Code and JetBrains

For developers who prefer a visual IDE, the Kilo Code extension can be installed directly from the VS Code Marketplace. For those in the JetBrains ecosystem, the Kilo Code plugin is available via the JetBrains Marketplace or through the in-IDE plugin settings. Both extensions provide inline autocomplete with ghost-text suggestions (using tab-to-accept), chat interfaces, and direct access to the specialized agent roster.

Command Line Interface (CLI)

For terminal-centric workflows, Kilo offers a robust CLI that can be installed across various package managers:

Advertisement

Shadow GPS — know where it is, always Real-time GPS tracking for vehicles, gear and loved ones. No monthly contracts.

Using npm, pnpm, or bun:

# npm
npm install -g @kilocode/cli

# pnpm
pnpm add -g @kilocode/cli

# bun
bun add -g @kilocode/cli

Using Homebrew (macOS and Linux) or Arch Linux (AUR):

# Homebrew
brew install Kilo-Org/tap/kilo

# Arch Linux
paru -S kilo-bin

Alternatively, developers can install Kilo via a direct curl script:

curl -fsSL https://kilo.ai/cli/install | bash

For environments without standard package managers, pre-built binaries are available for Windows (x64), macOS (Apple Silicon and Intel), and Linux (x64 and ARM). Notably, Kilo provides an x64-baseline build for older CPUs lacking AVX support, as well as a musl statically linked build optimized for minimal Docker images and Alpine Linux environments.

Autonomous Execution and CI/CD Integration

While interactive use inside an IDE is the most common workflow, Kilo also supports fully autonomous operations. By leveraging terminal and browser control, the agent can execute commands, run test suites, and even automate web-based tasks to verify its work.

This capability is particularly powerful when integrated into CI/CD pipelines. Using the CLI, developers can run Kilo in Autonomous Mode by appending the --auto flag:

kilo run --auto "run tests and fix any failures"

The --auto flag disables all interactive permission prompts, giving the agent full authority to execute actions, run commands, and modify files without manual confirmation. Because this mode allows unconstrained execution, it is highly recommended to run autonomous tasks only within secure, isolated, or trusted sandbox environments.

Expanding Capabilities with MCP

To prevent the AI from being isolated from external tools, Kilo integrates with the Model Context Protocol (MCP) marketplace. This allows developers to wire up various MCP servers directly to their agents, expanding their capabilities to interact with external databases, APIs, issue trackers, and proprietary internal tools.

For teams looking for off-host execution, Kilo also offers a web-based Cloud Agent, automated pull request reviews, and an always-on AI agent called KiloClaw. By combining open-source flexibility with deep environment integration, Kilo Code provides a highly adaptable foundation for teams looking to transition from basic AI autocomplete to fully realized agentic workflows.