Kilo Code

Kilo Code is an open-source AI coding agent shipped as a VS Code extension, a JetBrains plugin, and a CLI. It ships six built-in modes (Code / Architect / Debug / Ask / Orchestrator / Custom). TokenByte plugs in as its OpenAI Compatible provider.

Prerequisites

VS Code or a JetBrains IDE
The Kilo Code extension
A TokenByte API key from the console

Configuration

Open settings

On first install, click Use your own API key on the welcome screen.

Later on, press Cmd/Ctrl + Shift + P and run Kilo Code: Open Settings, or click the gear in the top-right of the extension sidebar.

Pick the provider

API Provider: OpenAI Compatible
Base URL: https://api.tokenbyte.ai/v1
API Key: paste your TokenByte key (starts with sk-tb-)
Model: enter the exact card title from the Models page or the Console → Models catalog

TokenByte also speaks Anthropic, but Kilo's Anthropic entry hits Anthropic directly with no base-URL override. To hit Claude through TokenByte, stay on OpenAI Compatible.

Optional: override model metadata

Expand Model Configuration to set context window, max output tokens, Vision / Computer Use, and pricing. Useful for custom models that TokenByte hasn't registered metadata for yet. Leave defaults for mainstream models.

Verify

Select some code in the editor, switch Kilo to Code or Ask mode, and try "Explain what this does". A useful reply means you're good.

All six modes (Code / Architect / Debug / Ask / Orchestrator / Custom) can bind a different model each, all sharing the same TokenByte key. In practice, Architect usually benefits from a planning/reasoning-focused model, while Code benefits from a faster coding-oriented model; use the exact card titles from the models catalog.

Stuck? See the FAQ.