Kilo Code is an open-source AI coding agent that runs in the same three places most developers actually work: VS Code, JetBrains IDEs, and a terminal CLI. It has climbed fast (past 22,000 stars as of 2026-06) in one of the most crowded categories on GitHub, where Cline, Continue, OpenCode, and a dozen others are all chasing the same job. Its angle is reach plus pricing: one agent across every surface, a catalog of 500+ models you can switch between mid-task, and what the project calls open pricing.

The pricing model is the real pitch

The phrase to understand is “pay the model provider’s rate with zero markup, no API keys required to start.” In practice that means you create a Kilo account and run models through it, and Kilo passes provider pricing through rather than reselling tokens at a margin. The friction it removes is real: you can start without wrangling your own OpenAI or Anthropic keys, and you can move between models without re-plumbing credentials.

The thing to keep clear is that the MIT-licensed agent in this repo and the billing path are two different layers. The extension and CLI are genuinely open source and you can point them at your own keys. The “no API keys, 500+ models” convenience runs through Kilo’s platform, which is also where the paid surfaces live: a cloud agent, automated PR code reviews, and an always-on agent product. This is an open-core project, and that is fine, but the README leads with the convenience and is quieter about which parts are the business.

What you actually get

Kilo ships task-specific modes you switch between: Code (the default, edits and writes code), Plan (designs and writes an implementation plan first), Ask (answers questions without touching files), Debug (traces issues), and Review (surfaces problems across performance, security, style, and tests). You can also define custom agents.

The capability list is the now-standard coding-agent kit: multi-file code generation from natural language, inline ghost-text autocomplete, self-checking where the agent reviews its own output, terminal and browser control, and an MCP marketplace for wiring up servers that extend what it can do.

Install

VS Code: install the Kilo Code extension from the Marketplace. CLI:

npm install -g @kilocode/cli
# or
curl -fsSL https://kilo.ai/cli/install | bash
# also: pnpm add -g @kilocode/cli, bun add -g @kilocode/cli,
# brew install Kilo-Org/tap/kilo, paru -S kilo-bin

Then run kilo in any project directory. JetBrains users install the Kilo Code plugin from the JetBrains Marketplace. Prebuilt CLI binaries are on the Releases page, including a musl static build for Alpine and minimal Docker images and an x64-baseline build for older CPUs without AVX.

Autonomous mode, and the caveat that comes with it

For CI/CD there is a fully autonomous mode:

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

The --auto flag disables every permission prompt and lets the agent execute any action without confirmation. That is genuinely useful in a pipeline and genuinely dangerous outside one. The README says to use it only in trusted environments, and you should take that literally: an agent with shell access and no confirmations can do real damage if pointed at the wrong repo or fed a hostile prompt. Sandbox it.

The honest read on pace and stability

Kilo moves fast, and the issue tracker shows what that costs. There are several hundred open issues at any time, which is partly a function of popularity and partly a function of shipping quickly. Two patterns are worth knowing before you commit a team to it.

First, updates can regress. One of the most-reacted issues in the tracker is a user revolt over a UI change in an update, which tells you the extension’s interface is still moving under users’ feet. If a stable, unchanging UI matters to you, pin a version and read release notes before upgrading.

Second, the surfaces are not equally mature. macOS-specific bugs and JetBrains plugin shell issues show up among the top issues, so the VS Code path is the most battle-tested and the JetBrains and CLI paths are newer. Pick the surface that is solid for your platform rather than assuming parity across all three.

Alternatives

ProjectSurfaceTake
Kilo-Org/kilocodeVS Code, JetBrains, CLIMulti-surface, 500+ models, open-core with platform billing
cline/clineVS CodeThe reference open-source VS Code agent; bring your own keys
continuedev/continueVS Code, JetBrainsHighly configurable, strong autocomplete heritage
anomalyco/opencodeCLI / terminalTerminal-native agent; the README says Kilo’s CLI forked from it

The README itself states that the Kilo CLI is a fork of OpenCode, so if you want the leaner terminal-only lineage, that is the one to compare against directly.

FAQ

Is Kilo Code free? The agent is open source under MIT and free to use with your own model API keys. The convenience path (an account, 500+ models, no keys) and the cloud and review products run through Kilo’s platform, which is where paid usage lives.

What is the difference between Kilo Code and Cline? Both are open-source AI coding agents. Kilo spans VS Code, JetBrains, and a CLI with an account-based model marketplace, while Cline is VS Code focused and bring-your-own-key. Kilo’s CLI is a fork of OpenCode per its README.

Can I run Kilo without an API key? Yes, through a Kilo account that routes 500+ models at provider pricing. You can also supply your own provider keys and skip the account if you prefer.

Is Kilo Code safe to run in CI? It has a kilo run --auto mode built for CI/CD that disables all permission prompts. Only run it in trusted, sandboxed environments, because it executes any action without confirmation.

Does Kilo work in JetBrains and on macOS reliably? There is a JetBrains plugin and macOS support, but top issues include JetBrains shell problems and macOS-specific bugs, so the VS Code surface is currently the most mature.