Chat2Graph - ScitiX

Chat2Graph is an open-source graph-native agent system from the TuGraph family. It calls large language models through LiteLLM, and LLM_ENDPOINT can point to any OpenAI-compatible service, so you can connect Chat2Graph to ScitiX Model Inference. This guide uses glm-5.2 as an example. For the full model list and pricing, see Models.

Configure

Create an API Key on the API Keys page. Copy .env.template to .env, then fill in:

MODEL_PLATFORM_TYPE=LITELLM
LLM_NAME=openai/glm-5.2
LLM_ENDPOINT=https://api.scitix.ai/model-api/v1
LLM_APIKEY=<YOUR_API_KEY>

EMBEDDING_MODEL_NAME=Qwen/Qwen3-Embedding-4B
EMBEDDING_MODEL_ENDPOINT=https://api.scitix.ai/model-api/v1/embeddings
EMBEDDING_MODEL_APIKEY=<YOUR_API_KEY>

Field	Description
`MODEL_PLATFORM_TYPE`	Set to `LITELLM`.
`LLM_NAME`	Use `openai/<model-id>` so LiteLLM treats the model as OpenAI-compatible and routes requests through `LLM_ENDPOINT`, such as `openai/glm-5.2`.
`LLM_ENDPOINT`	ScitiX OpenAI-compatible base URL with `/v1`: `https://api.scitix.ai/model-api/v1`
`LLM_APIKEY`	Your API Key created on the API Keys page
`EMBEDDING_MODEL_ENDPOINT`	ScitiX embedding endpoint: `https://api.scitix.ai/model-api/v1/embeddings`

Verify

Build and start Chat2Graph with ./bin/build.sh and ./bin/start.sh, open the Web UI at the default http://localhost:5010/, and send a message. A normal reply confirms that the connection works. You can also run a minimal LiteLLM check with the same configuration:

import litellm

r = litellm.completion(
    model="openai/glm-5.2",
    api_base="https://api.scitix.ai/model-api/v1",
    api_key="<YOUR_API_KEY>",
    messages=[{"role": "user", "content": "Reply with exactly: OK"}],
)
print(r.choices[0].message.content)

FAQ

LLM_NAME must include the openai/ prefix: This tells LiteLLM to treat the model as an OpenAI-compatible provider and use LLM_ENDPOINT instead of trying to infer another provider.
LLM_ENDPOINT is a base URL with /v1: Use https://api.scitix.ai/model-api/v1. The embedding endpoint uses the full /v1/embeddings path.
Model IDs: Use model IDs from the Models page.