Sentient Framework

Shows how the Sentient Framework fits into the overall system landscape, including external users, LLM services, vector databases, and code repositories.

Major containers and services: Service Layer (MCP Service, A2A Service), Agentic Core (Orchestrator, Tool Manager, Message Bus, Specialized Agents), and Context Engine Layer (Document/Graph/Keyword RAG, Hybrid Retriever, Context Compressor).

Detailed view of Service Layer: MCP endpoint (tool discovery, execution), A2A endpoint (task routing, agent discovery), tool registry, and agent card management.

The "brain" of the system: Orchestrator Agent (task decomposition, scheduling, synthesis), Message Bus, Tool Manager (validation, execution), Specialized Agents, and SentientPayload lifecycle.

Retrieval and optimization foundation: Document RAG (loader, splitter, embedder, vector store), Graph RAG (RepoMapper), Keyword RAG (BM25), Hybrid Retriever (RRF), and Context Compressor (EXIT-inspired).

Complete lifecycle of a complex query: ingestion, task decomposition, parallel retrieval, tool use, context compression, and response synthesis.

Production environment: Kubernetes cluster, Pod types (Service, Agent, Engine), managed services (Vector DB, Redis, PostgreSQL), external services (OpenAI, Gemini), and load balancing.

SentientPayload lifecycle: initial creation, state updates during execution, context fragment accumulation, compression, final output generation, and metadata enrichment.