Skip to main content

1. Start the server

API base URL: http://localhost:8080 On first up, the CLI downloads BGE-small-en-v1.5 (~67 MB, one-time) to ~/.moorcheh-edge/models. With --with-llm, it also pulls qwen2.5:0.5b-instruct (~400 MB) for moorcheh-edge answer.

2. Upload text documents (text store)

Save documents.json:
The CLI embeds text locally with BGE-small-en-v1.5 (384 dimensions) and uploads to a text store.

3. Search with plain text

4. Answer with RAG (local LLM)

Requires Ollama and the answer model. If you started with search-only up, run once:
After documents are uploaded:
Uses Ollama with qwen2.5:0.5b-instruct. Search-only (skip LLM setup):

5. Upload precomputed vectors (vector store)

To use your own embeddings instead, clear the store first (text and vector modes cannot mix):
Save vectors.json:

6. Python SDK

7. Stop (data is kept)

Data remains in ~/.moorcheh-edge/data.

Next steps

CLI

Full moorcheh-edge command reference

API reference

REST endpoint documentation

Python client

SDK workflow

Limits

10k store cap and dimension rules

Retail Kiosk example

PC + Arduino UNO Q in-store demo