Skip to main content

What is a Context?

PRISM Contexts are a primitive that consists of an arbitrary vector store database, and the supporting infrastructure to allow updates and modifications easily.

Because Contexts are decoupled from your agent, you can attach a single context to many agents! This allows you (or your organization) to load your own datasets into a single context and then reason about it in many different ways.

Context Types​

PRISM aims to support many different types of data for contexts. Each context is provisioned as a dedicated cluster, and is isolated from all other contexts in PRISM.

Different types of data are better reasoned about with different vector store types. These are the types of vector stores we either support today, or are on our roadmap:

Elastic Search​

PRISM Contexts are currently backed by Elastic Search clusters. These are the primary and default context type in PRISM.

Elastic Search clusters as a vector store are ideal for large amounts of generalized text data.

Common usecases include:

  • Website Content
  • Blog Content
  • Chat Messages

DragonflyDB (Redis Alternative)​

DragonflyDB Contexts are the opposite end of the spectrum from Elastic Search contexts. They provide access to context data that changes very frequently, but are able to store significantly less amounts of data.

DragonflyDB contexts as a vector store are ideal for financial data that changes quickly.

Common usecases include:

  • Solana Geyser Updates
  • DeFi Metrics
  • Model Outputs

Neo4J​

Neo4J is a graph database that allows for reasoning about the relationship between things. They provide access to context data that is highly relational and require reasoning about the complex relationships that often accompany financial transactions.

Neo4J contexts as a vector store are ideal for giving your agent access to complex/propreitary relationship graphs.

Common usecases include:

  • Predictive Analytics
  • Building Relationship Graphs