Trellis¶

Trellis is the home of Egeria's AI applications. Its egeria-trellis git repository holds the applications that use large language models (LLMs) and retrieval-augmented generation (RAG) alongside the open metadata ecosystem, together with the libraries they share.

Egeria is central to each Trellis application. It is the catalog of record for what the applications discover and create, and it is also the identity provider - you sign in to a Trellis application with your Egeria user identity, and the application calls Egeria as you. The elements it publishes record you as their owner. The applications call Egeria through pyegeria.

The applications¶

There are currently two applications:

• Resource Explorer discovers, surveys and catalogs information resources - git repositories, PostgreSQL databases and file systems - using Egeria as the catalog. It supports scouting (a fast, broad inventory across many resources), assessment (a deep analysis of a single resource's structure, security and quality), discovery (finding resources by what the surveys revealed) and enrichment (where people add context and answer the open questions that the surveys raised). Investigations are recorded in Egeria as projects with working sets of the resources being explored, and its survey results are stored as survey reports and annotations. It can also recover the architecture of the resources it has explored and, once a reviewer has accepted the findings, record it as solution blueprints.
• Egeria Advisor is a conversational assistant for people working with Egeria and pyegeria. It answers questions about Egeria's concepts and code, finds examples and definitions, and can carry out actions on your behalf through Dr.Egeria commands and reports. Its answers are tailored to the perspective you select, such as developer, data engineer, data steward or governance officer.

Each application is deployed and run independently of the other.

How Trellis is organized¶

The git repository is a single Python workspace (managed with uv) containing the two applications and six libraries that they share. The libraries cover authentication against Egeria (including the single sign-on hand-off from the Egeria Workspaces portal), a PostgreSQL/pgvector vector store, bounded context assembly for the LLM, the structure of ingested artifacts, step execution and query caching. Keeping the applications in one workspace means that this common code is shared without version skew, while each application remains a separately deployable service.

By default the applications use a local Ollama server for their language models, and a PostgreSQL database with the pgvector extension for their vector collections.

Running Trellis¶

There are two ways to run the applications:

• As an optional runtime alongside Egeria Workspaces. The trellis optional runtime starts Resource Explorer and Egeria Advisor against the Quickstart environment, and the Egeria Workspaces portal has tiles that open each application with single sign-on. Variants of the runtime use Ollama running on the host, in a container, or on an NVIDIA or AMD (ROCm) GPU.
• From a clone of the git repository, following its QUICKSTART.md. An Egeria platform is still needed, since it is the identity provider.

Raise an issue or comment below