25 editable reference diagrams
Generative AI Architecture Diagrams
Generative AI architecture diagrams show the systems around a model: retrieval, tool execution, document ingestion, permissions, evaluation and response delivery. Compare RAG, agent and serving patterns through their actual dependency paths.
Begin with the information a user is allowed to access and the actions a model may request. Add bounded tool permissions, source provenance, evaluation cases and fallback behavior. A prompt or vector index does not replace application authorization or output verification.
Choose a diagram for your design question
Open a reference to inspect its component flow and trade-offs, then adapt it in the browser-local editor. Manual editing does not require login; AI generation and account storage do.

RAG architecture diagram
Ground answers in authorized retrieved evidence.
8 components · 7 connections
Explore rag architecture diagram →
RAG LLM Production System Architecture
Add retrieval and source attribution
7 components · 8 connections
Explore rag llm production system architecture →
Enterprise RAG architecture
Preserve source permissions through retrieval.
9 components · 7 connections
Explore enterprise rag architecture →
Agentic RAG architecture
Bound tool loops and citation requirements.
6 components · 5 connections
Explore agentic rag architecture →
Multimodal RAG architecture
Retain source provenance across modalities.
8 components · 6 connections
Explore multimodal rag architecture →
Graph RAG architecture
Validate extracted relationships before treating them as facts.
7 components · 6 connections
Explore graph rag architecture →
Hybrid search RAG architecture
Calibrate fusion against evaluated queries.
7 components · 7 connections
Explore hybrid search rag architecture →
Permission aware RAG architecture
Filter before sending context to the model.
7 components · 6 connections
Explore permission aware rag architecture →
Multi tenant RAG architecture
Never trust a client supplied tenant identifier.
6 components · 5 connections
Explore multi tenant rag architecture →
Real time RAG architecture
Measure source-to-index freshness explicitly.
7 components · 6 connections
Explore real time rag architecture →
Document ingestion RAG architecture
Track document versions and deletion propagation.
7 components · 6 connections
Explore document ingestion rag architecture →
RAG evaluation architecture
Evaluate retrieval and groundedness separately.
6 components · 5 connections
Explore rag evaluation architecture →
RAG reranking architecture
Measure relevance gains against latency cost.
6 components · 5 connections
Explore rag reranking architecture →
Vector database semantic search architecture
Version embedding models and index metadata together.
7 components · 6 connections
Explore vector database semantic search architecture →
LLM application architecture
Keep authorization outside model instructions.
7 components · 6 connections
Explore llm application architecture →
LLM gateway architecture
Restrict provider access and spending centrally.
7 components · 6 connections
Explore llm gateway architecture →
LLM routing architecture
Route by measured quality and constraints.
6 components · 5 connections
Explore llm routing architecture →
LLM response caching architecture
Do not share sensitive answers across users.
7 components · 6 connections
Explore llm response caching architecture →
LLM guardrails architecture
Guardrails reduce risk without proving correctness.
7 components · 6 connections
Explore llm guardrails architecture →
LLM observability architecture
Exclude confidential prompts from default telemetry.
9 components · 6 connections
Explore llm observability architecture →
Multi agent AI architecture
Bound responsibilities and shared-state access.
8 components · 6 connections
Explore multi agent ai architecture →
AI agent tool calling architecture
The model proposes actions but policy authorizes them.
7 components · 6 connections
Explore ai agent tool calling architecture →
Human in the loop AI architecture
Do not execute consequential actions before approval.
6 components · 5 connections
Explore human in the loop ai architecture →
AI coding assistant architecture
Review patches before changing trusted code.
7 components · 6 connections
Explore ai coding assistant architecture →
Enterprise knowledge assistant architecture
Citations should identify accessible source versions.
8 components · 7 connections
Explore enterprise knowledge assistant architecture →