Guides/Agent integrations

Using wylon with agents

The wylon Token Factory exposes an inference API compatible with both OpenAI and Anthropic, so it can serve as the model backend for a wide range of coding agents (harnesses). Just point three values — Base URL, API Key, and model name — at wylon. No SDK swap or code rewrite required.

Pick a tool below for its configuration sub-page:

Tool	Form factor	Compatible protocol	Configuration
Claude Code	CLI	Anthropic-compatible	Environment variables
Codex	Desktop app	OpenAI-compatible	Config file
OpenCode	TUI	OpenAI-compatible	Config file / interactive
Openclaw	CLI / TUI / Dashboard	OpenAI-compatible	Interactive wizard
Hermes	CLI	OpenAI-compatible	Interactive wizard
WorkBuddy	Desktop app	OpenAI-compatible	Settings UI
DeepSeek Harness	CLI / Web UI	OpenAI-compatible	YAML / settings UI
TraeCode	Desktop app	OpenAI-compatible	Settings UI
Raccoon	Desktop app	OpenAI-compatible	Settings UI
Zcode	Desktop app	OpenAI-compatible	Settings UI
ComfyUI	Web UI	OpenAI Images-compatible	Custom node

Prerequisites

Before configuring any tool, prepare the following three pieces of information.

1. API Key

Sign in to the wylon console, then create and copy a key on the API Keys page (it looks like sk-...). Keep it safe — do not commit it to a repository or share it publicly.

key

Keep your key secureNever commit an API key to source control or include it in a client-side build. Use a secret manager or server-side environment variable in production.

2. Base URL (choose one, per the tool's protocol)

Protocol	Base URL	Notes
OpenAI-compatible	`https://api.wylon.cn/v1`	Used by most chat and coding tools
Anthropic-compatible	`https://api.wylon.cn`	Used by Claude Code; the client appends `/v1/messages` automatically, so do not include a path

3. Model name

wylon uses model IDs in the org/model form, for example:

A capable (high-quality) model, e.g. moonshotai/Kimi-K2.6
A lightweight (fast) model, e.g. deepseek-ai/DeepSeek-V4.1-Flash

lightbulb

These are examples only. For the complete, available model list, refer to the wylon Models page and substitute the exact model ID you intend to use.

FAQ

What if wylon is not listed as a provider?

Choose Custom / OpenAI-compatible and enter the configuration manually.

Most tools do not include a built-in wylon preset. Enter the Base URL, API Key, and model name manually. For ComfyUI, install an image-generation node that accepts a custom api_base.

Should the Base URL include /v1?

Include /v1 for OpenAI-compatible tools, but not for Claude Code.

OpenCode, Openclaw, Hermes, WorkBuddy, DeepSeek Harness, TraeCode, Raccoon, and Zcode use https://api.wylon.cn/v1. The example ComfyUI node uses https://api.wylon.cn/v1/. Claude Code uses https://api.wylon.cn because the client appends /v1/messages.

What should I enter as the model name?

Use the complete organization/model ID.

For example, moonshotai/Kimi-K2.6. See the Models page for current IDs.

Do not place API keys in code or configuration repositories.

Inject keys through environment variables whenever possible.

If a tool requires a local configuration file, make sure that file is excluded from version control.