AI Service Overview
How NQRust Analytics configures its LLM provider through config.yaml, and where to start.
The Analytics Service is the part of NQRust Analytics that turns plain-language questions into SQL. It does this by calling a large language model (LLM) and an embedding model, both of which you choose and configure. Because NQRust Analytics is self-hosted, you provide your own model access; there is no managed inference included.
How configuration works
All model settings live in a single file, config.yaml. This file controls:
- Which LLM generates SQL (the default in the installer is
gpt-4o-mini). - Which embedding model indexes your schema for retrieval.
- Which pipelines use which models, so you can mix and match per step.
Model access goes through a LiteLLM-style abstraction, which means any provider
that LiteLLM understands can be configured by editing config.yaml. You do not
need to write code for the common cases.
Ready-made templates
You do not have to author config.yaml from scratch. The installer ships
15 provider templates that you can copy and adjust:
OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Google AI Studio, Google Vertex AI, DeepSeek, xAI Grok, Groq, Qwen3, Zhipu, OpenRouter, Ollama, LM Studio, and a generic local-LLM template.
Select the template that matches your provider, enter your credentials and model names, and restart the service.
Where to go next
Use a Custom LLM or Embedder
Point NQRust Analytics at the model provider of your choice.
Customization
Add a provider that isn't covered by a template, and wire it into pipelines.
Good Practices
Tuning tips for latency, cost, and model selection.
Langfuse Setup
Trace and monitor your AI pipelines for observability.