Build RAG pipelines
you can debug.

Wire together retrieval nodes. Execute against real LLM APIs. Inspect data at every connection.

crag-pipeline.hachi

ready

Query"How does RAG work?"

Embeddingtext-embedding-3-small

Retrievertop_k: 5

docs: 5 · top_score: 0.94 · 156ms

Rerankercross-encoder

LLMgpt-4-turbo

5 nodes · 4 connections

latency: 245mspipeline healthy

Capabilities

Built for testing, debugging
and understanding RAG pipelines.

8 node types with typed connections. Drag, wire, configure.

query → embedding → retriever → reranker → llm

Click any connection. See embeddings, scores, reasoning.

score: 0.94 · docs: 5 · latency: 156ms

Your models, your vector store, your credentials. Live API calls, not a simulation.

openai / cohere / pinecone / qdrant

Whiteboard your architecture before you build it. Powered by Excalidraw.

plan → sketch → annotate → share

Live cursors, instant sync, shared results.

yjs-powered crdt

Org-scoped canvases, documents, credentials.

owner / admin / editor / viewer

hachi is free during beta.