Integration guides
SillyTavern
A locally self-hosted chat frontend focused on role-play and multi-turn conversation. It doesn't run models itself — it connects to a backend; choose the Custom (OpenAI-compatible) source under Chat Completion, enter the address and key, and you're on XylemNode.
- 01
Install and launch
SillyTavern runs on your machine (Node.js required). After following the official install guide, launch it on Windows by double-clicking Start.bat, or on macOS/Linux with ./start.sh; the official Launcher or Docker also work. Once running, open the local address shown in the terminal (default http://localhost:8000) in your browser.
- 02
Open API connections and choose the custom source
Click API Connections (the plug icon) at the top, switch the API type to Chat Completion, then set Chat Completion Source to Custom (OpenAI-compatible).
- 03
Enter the address and key
At the custom endpoint, enter XylemNode's address (with /v1) and your key.
- Custom Endpoint (Base URL)
https://api.xylemnode.com/v1- API Key
sk-your-key
- 04
Pick a model and connect
XylemNode implements /v1/models, so SillyTavern populates the model dropdown automatically — pick one (e.g. gpt-5.6-sol); if nothing loads, type the model ID in the text box. Click Connect, then use Test Message to send one and verify.
About prompt formatting
1) If a model has message-format requirements (e.g. only one system message, or strictly alternating roles) and errors out, expand Prompt Post-Processing and pick a preset like Merge consecutive messages or Semi-strict for compatibility. 2) When the endpoint is clearly fine but warnings keep popping up, tick Bypass API status check to skip the status check.
Common questions
SillyTavern says the model ID is not set, or errors that the system message must come first
The first is a pre-flight guard: you switched the model to manual entry and left the box empty, so the request body would carry an empty model string the upstream can't route, and it refuses to send. Enter the full model identifier, or go back to the model you picked from the dropdown. The second is a message-shape mismatch — errors like System message must be at the beginning have been reported: in Chat Completion mode ST spreads system instructions across several positions, while some models accept only a leading system message or require strictly alternating roles. Open Prompt Post-Processing and pick a preset such as Merge or Semi-strict so it folds the messages into a compliant shape before sending; no character-card changes needed.