--- title: "The RAG stack" type: "stack" slug: "rag-pipeline" canonical_url: "https://www.graphcanon.com/stacks/rag-pipeline" --- # The RAG stack Retrieval-augmented generation grounds an LLM in your own data. A production RAG pipeline is four layers: ingestion, a vector store, orchestration, and evaluation. ## 1. Data & Retrieval Ingestion & retrieval - parse, chunk, and load documents into a retrievable form. **When not to use:** Skip a heavy ingestion framework when your corpus is small and static; a script plus the embedding API is enough. **Category:** [Data & Retrieval](/categories/data-retrieval.md) - [generative-ai-for-beginners](/tools/microsoft-generative-ai-for-beginners.md) - 21 Lessons for Getting Started with Generative AI (★ 113,577) [Very active] - [graphify](/tools/graphify-labs-graphify.md) - Turn any code or documentation into a queryable knowledge graph (★ 107,507) [Very active] - [awesome-llm-apps](/tools/shubhamsaboo-awesome-llm-apps.md) - Over 100 runnable AI Agent and RAG apps to clone, tweak, and deploy. (★ 131,230) [Very active] - [ragflow](/tools/infiniflow-ragflow.md) - Retrieval-Augmented Generation engine with agent capabilities (★ 86,541) [Very active] ## 2. Vector Databases Vector store - persist embeddings and run similarity search at query time. **When not to use:** Don't reach for a dedicated vector DB under ~100k vectors; pgvector on your existing Postgres is simpler to operate. **Compare:** - [Qdrant vs Chroma](/compare/chroma-core-chroma-vs-qdrant-qdrant.md) - [Qdrant vs pgvector](/compare/pgvector-pgvector-vs-qdrant-qdrant.md) - [Qdrant vs Milvus](/compare/milvus-io-milvus-vs-qdrant-qdrant.md) **Category:** [Vector Databases](/categories/vector-databases.md) - [milvus](/tools/milvus-io-milvus.md) - High-performance cloud-native vector database (★ 45,402) [Very active] - [WeKnora](/tools/tencent-weknora.md) - Open-source LLM knowledge platform for creating a queryable RAG, autonomous reasoning agent, and self-maintaining Wiki. (★ 19,992) [Very active] - [qdrant](/tools/qdrant-qdrant.md) - High-performance, massive-scale Vector Database and Vector Search Engine (★ 33,629) [Very active] - [tidb](/tools/pingcap-tidb.md) - Scalable, cloud-native database with ACID transactions and vector search support. (★ 40,446) [Very active] ## 3. LLM Frameworks Orchestration - assemble retrieval, prompting, and the model into a chain. **When not to use:** Avoid a framework for a single prompt-and-retrieve call; the abstraction can cost more than it saves. **Compare:** - [LangChain vs LlamaIndex](/compare/langchain-ai-langchain-vs-run-llama-llama-index.md) - [Haystack vs LangChain](/compare/deepset-ai-haystack-vs-langchain-ai-langchain.md) **Category:** [LLM Frameworks](/categories/llm-frameworks.md) - [AutoGPT](/tools/significant-gravitas-autogpt.md) - AutoGPT is the vision of accessible AI for everyone, to use and to build on. (★ 186,623) [Very active] - [transformers](/tools/huggingface-transformers.md) - Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models (★ 164,121) [Very active] - [tensorflow](/tools/tensorflow-tensorflow.md) - An Open Source Machine Learning Framework for Everyone (★ 196,758) [Very active] - [prompts.chat](/tools/f-prompts-chat.md) - Share, discover, and collect prompts from the community (★ 167,212) [Very active] ## 4. Evaluation & Observability Evaluation & tracing - measure answer quality, cost, and latency before and after changes. **When not to use:** Defer heavyweight eval infra only until you have real traffic - never skip it once users depend on answers. **Category:** [Evaluation & Observability](/categories/evaluation-observability.md) - [generative-ai-for-beginners](/tools/microsoft-generative-ai-for-beginners.md) - 21 Lessons for Getting Started with Generative AI (★ 113,577) [Very active] - [headroom](/tools/headroomlabs-ai-headroom.md) - Compress tool outputs and data to reduce tokens before reaching the LLM. (★ 66,470) [Very active] - [CL4R1T4S](/tools/elder-plinius-cl4r1t4s.md) - Leaked system prompts for various AI agents include ChatGPT, Claude, Gemini among others emphasizing transparency and access. (★ 46,435) [Very active] - [LibreChat](/tools/danny-avila-librechat.md) - Enhanced ChatGPT Clone with extensive features and integrations for self-hosting (★ 41,282) [Very active] ## Common questions ### What is the the rag stack? Retrieval-augmented generation grounds an LLM in your own data. A production RAG pipeline is four layers: ingestion, a vector store, orchestration, and evaluation. ### When should I use the the rag stack? Use this stack when your constraints match its layers: Data & Retrieval, Vector Databases, LLM Frameworks, Evaluation & Observability. Each step on the page includes when-not-to-use guidance so you do not over-engineer. ### How does GraphCanon pick tools for the rag stack? Stack pages are editorial workflows over canonical categories. Representative tools are pulled live from the graph at render time (top adoption in each category), not a fixed marketing list. ### Where can I compare tools in this stack? Follow category hubs and head-to-head compare pages linked from each step. Featured comparisons cover the highest-intent pairs (frameworks, vector DBs, agent runtimes). Start at [compare hub](/compare). ### Is there a machine-readable version of The RAG stack? Yes. Append `.md` to [`/stacks/rag-pipeline`](/stacks/rag-pipeline) or fetch [`/md/stacks/rag-pipeline`](/md/stacks/rag-pipeline) for a markdown twin with steps, when-not notes, and tool links. --- **Machine-readable endpoints** - JSON: [`/api/graphcanon/categories`](/api/graphcanon/categories) - LLM index: [/llms.txt](/llms.txt) - Full corpus: [/llms-full.txt](/llms-full.txt) _GraphCanon - The knowledge graph for AI development. https://www.graphcanon.com/_