Moorcheh On-PremOwn your data. Run your search. Keep your memory.
Self-hosted semantic search and RAG - free Community Edition, source-available on GitHub, local by default. Choose Ollama, OpenAI, or Cohere for embeddings and LLM answers. Your index stays under ~/.moorcheh/data.
You Should Own Your Data
Not as a slogan - as architecture.
Your documents, embeddings, and queries belong on your machine - not spread across a vector SaaS, reranker API, and glue code you do not control.
Index hosted by a vector SaaS
Index under ~/.moorcheh/data
Index under ~/.moorcheh/data
Queries logged by a third party
Queries stay on localhost
Queries stay on localhost
Separate reranker API
Built-in ITS scoring
Built-in ITS scoring
Five services to wire together
moorcheh up
moorcheh up
Need billion-scale in a private VPC? Moorcheh Enterprise
Your Machine. Your Pipeline.
Text and vector namespaces follow different paths - both stay entirely local.
One command. Full stack.
moorcheh up starts the server in Docker, mounts ~/.moorcheh/data, and runs setup - choose Ollama, OpenAI, or Cohere for embeddings and LLM.
What Makes Moorcheh Different
Three layers - compress at ingest, score at query, rank without a reranker.
Not pgvector on Postgres - a purpose-built information-theoretic retrieval engine. Same stack on-prem as cloud.
- MIBAt ingest
Maximum Information Binarization
- EDMAt query
Efficient Distance Metric
- ITSBuilt-in
Information-Theoretic Scoring
Cards auto-rotate · hover to pause
Maximum Information Binarization
Float embeddings become compact one-bit codes - ~96 B per 768-d vector vs ~3 KB float32.
Efficient Distance Metric
CPU-friendly bitwise scoring instead of high-dimensional float cosine - no HNSW graph in RAM.
Information-Theoretic Scoring
Replaces the separate reranker pass - no cross-encoder model or second vendor API to wire in.
Built for Developers Who Own Their Stack
Self-hosted retrieval and RAG - same engine as cloud, on your machine.
Your data stays yours
Documents, embeddings, and queries on localhost - index under ~/.moorcheh/data.
Source-available on GitHub
Moorcheh Community License - free for single-node, non-commercial use. Enterprise for commercial and multi-node.
Ollama, OpenAI, or Cohere
Configure embedding and LLM providers with moorcheh configure - fully local with Ollama, or cloud models while your index stays on-prem.
Built-in scoring
Information-theoretic retrieval - no separate reranker API to wire in.
Docker does the work
moorcheh up pulls moorcheh/server, mounts your data, and exposes localhost:8080.
Persistent storage
Data survives moorcheh down. Back up ~/.moorcheh anytime.
Built-In Limits
Transparent quotas so you know exactly what to expect on-prem.
No cloud-tier cap
Create as many text or vector namespaces as you need - no cloud-tier namespace count limit.
Global storage quota
100,000 items total across all namespaces (text documents and vectors combined). Check usage via /health or moorcheh status.
Embed & LLM choice
Ollama, OpenAI, or Cohere for text embeddings and /answer generation - set once in ~/.moorcheh/config.json via moorcheh configure.
Start in 5 Minutespip install moorcheh-client · moorcheh up · search locally
First run configures Ollama, OpenAI, or Cohere for embeddings and LLM via moorcheh configure.