Skip to main content

​
What is Moorcheh Edge?

Moorcheh Edge runs locally on your machine for fast vector upload and similarity search. The client sends float embedding vectors; the server binarizes them and searches on-device without cloud calls.
Edge has no API keys. Start the server with moorcheh-edge up, then call the API on localhost.

​
Store modes

One flat store per container - no namespaces. You cannot mix text and vector modes in the same store. Use moorcheh-edge clear-store to reset.

​
What you can do

  • Upload text documents (client embeds locally) or precomputed vectors
  • Search with a text query (CLI/SDK embed locally) or a float query vector
  • Answer questions (RAG) - search + local LLM via Ollama (qwen2.5:0.5b-instruct); stream tokens with POST /answer/stream
  • Voice on edge hardware (Linux) - STT/TTS and moorcheh-edge voice serve for kiosk mic/speaker (see Voice CLI)
  • Delete items by id or wipe the entire store
  • Export catalog text - GET /export returns all chunk text (no vectors) for sync to another app
  • Check store health and quota via /health

​
Prerequisites

  • Python 3.10+
  • Docker (Desktop or Engine)
No Ollama required for search. For answer / RAG, run moorcheh-edge up --with-llm on Linux - it installs Ollama and pulls qwen2.5:0.5b-instruct (~400 MB). Text embedding uses FastEmbed (BAAI/bge-small-en-v1.5, 384 dimensions) bundled with the Python package. When search returns no matching passages, /answer and /answer/stream return a fixed I don’t have enough information to answer that question. without calling the LLM.

​
Install and run

See Quickstart for upload, search, and answer.

​
Documentation

​
Data persistence

Items are stored under:
moorcheh-edge down stops the container but does not delete this folder.