Agentics & Retrieval

The AI-flavoured capabilities. Both are M2 — narrow surface still settling. Use for prototypes / non-critical workloads; expect minor breaking changes within preview minors.

Packages

Click any package name to jump to its source-doc.

Package	NuGet ID	Maturity	Capability
Agentics	Cephalon.Agentics	M2	Capability.Agentics
Retrieval	Cephalon.Retrieval	M2	Capability.Retrieval

Cephalon.Agentics

Agentic workload runtime — task graph, tool registry, scheduled agents.

What it ships

• IAgentToolDispatcher — invoke registered tools from an agent run.
• IAgentToolRunCatalog — list/query tool runs (for observability).
• AgentToolExecutionRequest / Result / Outcome — typed envelopes.
• Source generator for [AgentTool]-tagged methods → automatic registry registration.

Use cases

Use case	Pattern
LLM with tool-use	Register tools with [AgentTool]; dispatcher invokes typed methods
Scheduled agent runs	Cephalon.Worker + Cephalon.Agentics
Multi-step workflow with checkpoints	Process manager (Cephalon.Behaviors.Patterns) + tool calls

Configuration

appsettings.json

{
  "Engine": {
    "Agentics": {
      "Enabled": true,
      "RunHistory": { "Retention": "30.00:00:00" },
      "Concurrency": 4
    }
  }
}

Cephalon.Retrieval

Retrieval / RAG runtime — vector search, hybrid retrieval, ranking.

What it ships

• IRetrievalQuery — typed retrieval interface.
• IEmbeddingProvider — pluggable embedding generation (no built-in provider; bring your own).
• Vector store integration via Cephalon.Data.Qdrant.

Use cases

Use case	Pattern
RAG over docs	Embed documents → store in Qdrant → IRetrievalQuery
Hybrid keyword + semantic	Elasticsearch + Qdrant; engine ranks results
Reranking	Plug a IRetrievalReranker after initial retrieval

Configuration

appsettings.json

{
  "Engine": {
    "Retrieval": {
      "Enabled": true,
      "VectorStore": "Qdrant",
      "Embeddings": { "Provider": "OpenAI" },     // BYO via IEmbeddingProvider
      "Reranking": { "Enabled": false }
    }
  }
}

Dependencies

Cephalon.Retrieval  needs  Cephalon.Data + Cephalon.Data.Qdrant
                    needs  an IEmbeddingProvider implementation (not shipped — you provide)

See also

• Technology → Agentics — narrative.
• Technology → Retrieval — narrative.
• Tutorial → Agentic workload — walkthrough.
• Tutorial → Retrieval / RAG — walkthrough.