Glossary
RAG Source
A RAG source is an external data repository that an AI agent retrieves information from at inference time using Retrieval-Augmented Generation, directly influencing the accuracy and currency of the agent's responses.
What is RAG Source?
Retrieval-Augmented Generation (RAG) is a technique where an AI agent supplements its foundation model's knowledge by retrieving relevant information from external data sources at the time it generates a response. RAG sources can include document databases, knowledge bases, APIs, vector stores, and any other data repository the agent can query. The quality, currency, and relevance of these sources directly affect the agent's output quality.
RAG sources are a critical trust factor because they determine the factual basis of an agent's responses. An agent retrieving from a curated, up-to-date, authoritative data source will produce more accurate outputs than one drawing from stale, incomplete, or unreliable sources. The trust implications extend beyond accuracy -- using inappropriate RAG sources can introduce bias, include copyrighted material, or surface confidential information.
From a scoring perspective, RAG sources must be tracked as a configuration component. Changes to RAG sources -- adding new sources, removing existing ones, or updating the underlying data -- alter the agent's knowledge base and therefore its behavior. RAG source quality assessment is an emerging area of agent evaluation, considering factors like data freshness, source authority, coverage breadth, and retrieval accuracy.
Example
A legal research agent uses three RAG sources: (1) a vector database of federal court opinions updated weekly, (2) a statutory code database updated monthly after legislative sessions, and (3) a curated collection of law review articles updated quarterly. When answering a legal question, the agent retrieves relevant passages from all three sources and synthesizes them with its model's reasoning capability. If the statutory database falls behind after a major legislative session, the agent's answers about new laws will be incorrect until the source is updated.
How Signet addresses this
Signet includes RAG source configuration in the agent's configuration fingerprint. Changes to RAG sources are classified as tool/data changes and trigger appropriate score decay. Signet's Quality dimension evaluates the impact of RAG sources on output accuracy, and the Stability dimension considers how frequently RAG sources are updated or changed. Operators can document their RAG source update cadence in their agent's profile to demonstrate data currency practices.
Build trust into your agents
Register your agents with Signet to receive a permanent identity and trust score.