The World's First Information-Theoretic Search Engine

Ultra-fast, deterministic, index-free semantic search — and the full AI infrastructure stack that grew around it.

Moorcheh pioneered the information-theoretic approach to semantic search — the same foundational direction Google Research later brought to mainstream attention with their TurboQuant paper. We've been shipping production systems on these principles for two years, and we've gone well beyond what the academic papers describe.

While TurboQuant and RaBitQ address compression, Moorcheh combines three integrated components that operate as a whole:

• MIB — Maximally Informative Binarization · information-theoretic compression tuned to preserve semantic content, not just minimize geometric distortion.
• EDM — Efficient Distance Metric · bitwise operations on native CPU instructions, replacing the floating-point math behind cosine similarity.
• ITS — Information-Theoretic Score · a ranking mechanism that replaces geometric proximity as the relevance signal, with an integrated reranker. This is the part the papers don't touch — and it's why our end-to-end latency beats two-stage (search + external rerank) architectures by ~6.6×.

Compression alone can't get you there. Compression plus the right scoring function plus the right execution model can.

📄 Moorcheh White Paper: arxiv.org/abs/2601.11557 · 📰 Our take on the TurboQuant moment: Read the blog

Benchmarked across 14 MAIR datasets and 10,038 queries — constant throughput across all precision levels, something HNSW-based systems structurally cannot do.

Out of the engine grew a fully serverless, cloud-native architecture that powers large enterprise customers — saving millions while maintaining production-grade AI infrastructure and knowledge management.

Moorcheh introduced the first index-free database architecture: no dependence on RAM, no delay from indexing and re-indexing. It performs a full scan of every entry in a few milliseconds — deterministic, exact-match retrieval at any scale.

Real impact: one enterprise deployment went from $2.5M/year → $36K/year in retrieval infrastructure costs.

• ✅ Sovereign by design — deploys into your VPC on AWS, GCP, or Azure
• ✅ Zero egress — no data ever crosses your perimeter
• ✅ True serverless economics — scales to millions of queries or to $0 when idle
• ✅ Deploys in under 10 minutes via Infrastructure-as-Code

Everything below grew out of the core engine — a complete, open, and community-driven AI stack.

A complete studio for testing, building, and deploying AI applications and chatbots — with free credits, a full playground, and a comprehensive set of functional API endpoints supporting the most complex workflows.

A companion memory agent that pairs with any working agent to manage memories across sessions, tools, and time. Fully open source and community-driven, it inherits Moorcheh's deterministic retrieval — eliminating the behavioral drift that ANN-based memory introduces into multi-turn agents. Continually gaining traction with AI developers and agent builders.

📄 Memanto White Paper: arxiv.org/abs/2604.22085 · 🌐 memanto.ai · 📰 Memanto Blog · 📈 Download stats

Born from the need for higher accuracy and functionality in OpenClaw / NeMo Claw architectures. Runs entirely through the Moorcheh API ecosystem and a single API key — no external connectors required. The only known method of running and deploying OpenClaw bots fully contained within an air-gapped infrastructure.

A free on-premise version that pairs with local models and performs every cloud functionality directly from your own machine — zero cloud reliance.

The newest addition: a lightweight, containerized search engine that runs on even the most restricted devices at incredible speed and accuracy.

Many community-driven projects help indie developers and hackers tap into Moorcheh's engine — including the JS Chat Boilerplate, which lets you ship a full chat application with minimal code.

Moorcheh plugs natively into the major AI infrastructure and agentic stacks:

LangChain · LangGraph · LlamaIndex · n8n · CrewAI · Hugging Face · MCP · and more.

Install the SDK:

pip install moorcheh-sdk

Build a RAG system in seconds:

import os
from moorcheh_sdk import MoorchehClient

client = MoorchehClient(api_key=os.getenv("MOORCHEH_API_KEY"))

# Create a namespace and ingest documents
client.create_namespace("my-rag", "text")
client.upload_documents("my-rag", [
    {"id": "doc1", "text": "Your content...", "metadata": {}}
])

# Get an AI-powered, grounded answer
answer = client.get_generative_answer(
    namespace="my-rag",
    query="Your question here"
)
print(answer["answer"])

Repo	⭐ Stars	Description
memanto		🐜 Memory that AI Agents Love — the companion memory agent
moorcheh-python-sdk		🐍 Official Python SDK for the Moorcheh Semantic Search API
moorcheh-chat-boilerplate		💬 Production-ready Next.js chat app powered by ITS search + RAG
moorcheh-mcp		🔌 MCP server bringing Moorcheh search & RAG to MCP clients
memantoclaw		🦅 Air-gapped OpenClaw deployment via the Moorcheh API
langchain-moorcheh		🦜 Seamless LangChain ↔ Moorcheh integration
n8n-nodes-moorcheh		⚡ n8n nodes for ingestion, semantic search & AI workflows
moorcheh-examples		📓 Practical sample projects and integration recipes
moorcheh-benchmarks		📊 Benchmarks, experiments, and use cases

➡️ Browse all repositories →

Real, runnable examples across regulated and document-heavy domains:

• 🌐 Website: moorcheh.ai
• 📝 Moorcheh Blog: moorcheh.ai/blog
• 🐜 Memanto: memanto.ai · memanto.ai/blog
• 📄 Moorcheh Paper: arxiv.org/abs/2601.11557
• 📄 Memanto Paper: arxiv.org/abs/2604.22085
• 📚 Docs: docs.moorcheh.ai