Integration guides
Coze Studio (open-source)
The open-source, self-hosted edition of Coze for visually building AI agents and workflows. Its model service supports 'third-party API relay' — choose the openai protocol and enter a /v1 address to connect XylemNode.
- 01
Deploy the open-source Coze Studio
The Coze cloud platform (coze.cn / coze.com) uses platform-hosted models and can't connect custom endpoints; the one that connects XylemNode is the open-source Coze Studio. Following the official docs, clone the repo and deploy it with Docker Compose.
- 02
Open Model Management
In your browser, open the Model Management page in the admin console and click Add Model.
- Model management URL
http://localhost:8888/admin/#model-management
- 03
Fill in as a third-party relay
XylemNode counts as 'other third-party API relay': choose the openai protocol, enter the domain with /v1 as the address (not /chat/completions), then add the key and model name.
- Protocol
openai- Base URL
https://api.xylemnode.com/v1- API Key
sk-your-key- Model name
gpt-5.6-sol
- 04
Save and use
After saving, the model ID is globally unique — don't change it once live (doing so breaks calls). From then on you can select this model when creating agents or workflows.
Two notes
1) Only the open-source Coze Studio supports this kind of connection; the Coze cloud platform doesn't allow custom OpenAI-compatible endpoints. 2) If you deploy with Docker and the model service runs on the host, localhost inside the container isn't the host — use the host IP or host.docker.internal; a public address like XylemNode's isn't affected.
Common questions
The model dropdown is empty when I create an agent in Coze Studio
The open-source edition ships with no usable models, and the documentation is explicit that you must configure a model service after deploying or you won't be able to pick one while creating an agent or workflow. Add one at /admin/#model-management: protocol openai, and an endpoint that stops at /v1 rather than /chat/completions — which is exactly what the docs prescribe for third-party API relays. Two official cautions are worth reading first: don't change a model ID once it is live, because agents and workflows call models by ID and a change breaks those calls; and before deleting a model, confirm no live traffic still depends on it.