Agent Framework RAG (TextSearchProvider)

ConnectSoft’s ConnectSoft.Extensions.AI.AgentsFramework library can register agents with TextSearchProvider, an AIContextProvider that retrieves text before (or on-demand around) each model call.

Layers

Layer	Type	Purpose
Application retrieval	IRetrievalService	Semantic search over Microsoft.Extensions.VectorData (see AI extensions).
Agent Framework	TextSearchProvider	Injects retrieved passages into the agent pipeline via ChatClientAgentOptions.AIContextProviders.
Bridge	IAgentTextSearchAdapter	Maps a query string to TextSearchProvider.TextSearchResult (implemented in BaseTemplate ConnectSoft.BaseTemplate.AgentFramework with RetrievalAgentTextSearchAdapter → IRetrievalService).

Configure per-agent AgentRegistrationOptions.Rag (Enabled, VectorCollectionName, TopK, SearchTime, CitationsPrompt, UseInMemoryChatHistory).

Alignment with ingestion

Use the same collection name for:

• AgentFramework:Agents[*]:Rag:VectorCollectionName
• Ingestion / MicrosoftExtensionsAI:VectorStore:CollectionName (and pipeline targets)

so agents search the corpus produced by vector ingestion.

Tutorials

• Jamie Maguire — RAG with TextSearchProvider and in-memory vector store
• Microsoft Agent Framework overview

Agent Skills & harness (related)

• Agent Skills — folder-based SKILL.md libraries via FileAgentSkillsProvider (optional AgentFramework:AgentSkills in ConnectSoft).
• Agent Harness — advanced execution/hosting; not part of the default BaseTemplate surface; evaluate risk before enabling.