Pipeline Stages
📄 Data Source
Files, URLs, or inline text to index
✓
✂️ Chunking
Split documents into retrievable chunks
✓
🔢 Embedding
Convert chunks to vectors
✓
🔍 Retrieval
Dense, sparse, or hybrid search
🏆 Ranking
LLM or cross-encoder re-ranking
✨ Generation
Model, prompt, and self-correction config
🛡️ Guardrails
PII detection, cost cap, content filters
MeshFlow RAG supports HybridRetriever (BM25 + dense), LLMRanker, SelfCorrectingRAG, and cache_control prompt caching — up to 90% cost reduction.

Pipeline Preview

📄 Data Source
files: [] · urls: [] · inline: —
✓
↓
✂️ Chunking
chunk_size: 500 · overlap: 50 · character
✓
↓
🔢 Embedding
provider: builtin · dims: 384
✓
↓
🔍 Retrieval
strategy: dense · top_k: 5 · max_chars: 4000
!
↓
🏆 Ranking
ranker: none (skip)
!
↓
✨ Generation
model: claude-sonnet-4-6 · self-correct: off · cache: on
!
↓
🛡️ Guardrails
none configured
!
Configure: Data Source