contenox

Kilo Code: what it is, who owns it, and the alternatives

Updated

COMPARE Kilo Code

Kilo Code is an open-source AI coding agent (MIT) that runs inside VS Code and JetBrains, in the terminal as a CLI, in a desktop app in beta, and in the cloud, with any model: your own keys, local models through Ollama or LM Studio, or Kilo Gateway, its own endpoint to hundreds of models including free ones. Anaconda bought it on 15 July 2026; the license, the free tier and the products stayed, and Anaconda added organization-level governance on top. It is not from Amazon, which is a common mix-up: Amazon’s teams use it, and it connects to Bedrock, and that’s the whole link.

What Kilo Code is

Kilo Code
MakerAnaconda, since July 2026; before that Kilo-Org, which forked it from Roo Code
LicenseMIT
InterfacesVS Code extension, JetBrains plugin, CLI (npm install -g @kilocode/cli), Kilo Desktop (beta), cloud agents
ModelsKilo Gateway with 500+ models and a free-models option; your own keys for any provider; Ollama and LM Studio
SubscriptionsChatGPT plans and Kilo’s own plans; not Claude plans
PermissionsAllow, ask or deny per tool; asks by default; auto-approve by category
SandboxmacOS and Linux; not Windows
ModesCode, Architect, Debug and Orchestrator, each with its own prompt and tool set
TeamsShared sessions, code reviews, usage tracking; organization policy and model blocking after the Anaconda deal

The free-models option is the feature people come for: pick “free” in the model list and Kilo routes to whichever models currently cost nothing on the gateway, with no key and no card. The catch is in the word “currently”; the list changes, and the free models are not the strong ones.

What changed with Anaconda

Anaconda’s stated reason was the enterprise gap: companies couldn’t see what their developers spent on AI, which models saw their code, or stop them depending on one vendor. What it added:

  • One view of AI activity across VS Code, JetBrains and the CLI, for an organization.
  • Policy enforcement and model routing. If an organization blocks a model for a team, it disappears from that team’s model list.
  • Nothing removed. Anaconda’s post says Kilo stays available to individuals, teams and organizations, and keeps its open-source commitment. The repository is still MIT.

For an individual developer, the day after the deal looked like the day before. For a company already on Anaconda for Python, it turned Kilo into the agent that comes with the governance they have.

Where Kilo Code came from

Kilo Code is a fork of Roo Code, which was itself a fork of Cline. Roo Code was archived on 15 May 2026; its README points users to Cline and a community fork, and most of its users went to Kilo Code or back to Cline. That history is why Kilo and Cline look alike: the same modes, the same auto-approve categories, the same .rules idea. The differences are the gateway, the free models, the sandbox and now the owner.

Kilo Code alternatives

ToolWhat it isModelsPick it if
ClineThe agent Kilo descends from: VS Code, JetBrains, CLI, desktop app; Apache-2.0Many providers with your keys; ChatGPT sign-inYou want the same editor agent without a gateway in the middle
OpenCodeOpen-source terminal agent (MIT), desktop app in beta75+ providers; ChatGPT and Copilot sign-in; localYou’d rather work in the terminal than the editor
CodexOpenAI’s open-source agent with a native Windows sandboxOpenAI with a ChatGPT plan; local with --ossYou pay for ChatGPT and you’re on Windows
CursorAn AI editor with the best tab completion and an agentCursor-hosted modelsYou want the editor experience more than model choice
ContenoxOpen-source desktop app with a policy engine for what the agent may runChatGPT plan, API keys, Ollama, vLLM; hosts Claude CodeYou want the rules visible and kept, on Windows, outside the editor

Kilo’s sandbox is a real advantage over Cline and OpenCode on macOS and Linux, and no advantage on Windows, where it doesn’t exist. How to sandbox AI coding agents covers what each agent offers there.

Contenox

Contenox is for the part Kilo’s auto-approve categories decide loosely: which commands run without asking. Instead of a category switch, a rule list you can read in Permissions, per project, with builds and tests allowed and risky commands asking. It runs with the same providers, including local models, and hosts Claude Code when you have a Claude plan. Download it here.

Related

Esc to close