Configuring External LLM Services
10/09/2026
This page details the LLM integration framework in Graph Modeling (formerly PoolParty) version 10.1 and higher, outlining how to use system properties to connect diverse model providers to your taxonomy development environment.
Earlier versions of Graph Modeling featuring the Extend your Taxonomy feature were limited to using Amazon Bedrock for all LLM capabilities. While this ensured stability, it restricted customers to Bedrock-specific models and prevented integration with other providers or self-hosted options.
With the version 10.1 LLMs become pluggable runtime components. Graph Modeling supports three broad model types:
Cloud proprietary LLMs such as OpenAI, Anthropic, Google, Cohere, offering strong reasoning and reliability.
Open-source/self-hosted models such as Llama, Mistral, Qwen, Phi, suitable for full data control, private deployments, and fine-tuning.
Hybrid inference providers: services like TogetherAI or OctoAI that host optimized variants of open models.
This approach decouples model choice from the platform, allowing teams to adopt the LLM strategy that best fits performance, compliance, and cost requirements.
Note
An active LLM configuration is required to utilize all available Taxonomy Builder features. Graphwise provides a paid professional service to handle this setup for customers, if required. For pricing and other details, please reach out to your Graphwise representative.
Important
The availability of specific LLM versions depends directly on the supported LangChain4j version in Graph Modeling:
Graph Modeling 10.3 and prior:
langchain4jversion 1.12.2Graph Modeling 10.4:
langchain4jversion 1.19.0
If you are attempting to configure newer models that are unavailable in your current setup, check the LangChain4j Documentation to verify model version compatibility for your specific release.