Technical Documentation

What is REM Network?

The REM Network is a Decentralized Physical Infrastructure Network (DePIN) that provides vector database services for AI applications. We're building a decentralized alternative to Pinecone, powered by Proof of RAM consensus and the Sui blockchain.

Key Features

• Decentralized - No single point of failure
• Fast - Target p99 latency ≤ 500ms for 1M vectors
• Cost-Effective - Market-driven pricing via DePIN economics
• Transparent - Blockchain-verified operations

Proof of RAM (PoRAM)

Our novel consensus mechanism that cryptographically proves miners actually possess the RAM they claim.

How It Works

1. Commitment Phase
   • Miner claims N GB of available RAM
   • Coordinator generates epoch seed (rotates every hour)
   • Miner generates deterministic data in RAM based on seed
2. Challenge Phase
   • Coordinator sends random offset challenges: (offset, chunk_size)
   • Miner must return: hash(data[offset:offset+chunk_size]) + response time
   • Challenges are unpredictable and cannot be pre-computed
3. Verification Phase
   • Coordinator verifies hash matches expected value
   • Measures latency (RAM access should be <100μs)
   • Repeated failures → miner marked as dishonest

Reputation System

REM Network uses a dynamic reputation system. Your score determines your reward multiplier — built from real performance metrics, not just hardware.

Tier	Score Range	Multiplier	Description
Elite	90-100	1.5×	Premium hardware, perfect uptime
Trusted	70-89	1.0×	Reliable, good performance
Standard	50-69	0.75×	Average performance
Probation	30-49	0.25×	New or inconsistent
Restricted	0-29	0×	SSD scammers, auto-slashed

Score Components

• RAM Performance (40%): Challenge latency per GB — proves real RAM
• Reliability (25%): Challenge pass rate + uptime + consecutive online streak
• Query Performance (20%): Response time vs network median
• Network Quality (15%): Heartbeat regularity + WebSocket latency

Reward Formula

Miners earn rewards based on three components:

Miner Rewards = 
    [Capacity Score × 50%] +
    [Challenge Score × 40%] +
    [Query Score × 10%]
    
    × Reputation Multiplier
    × Uptime²

• Capacity Score: Your RAM / Total Network RAM
• Challenge Score: PoRAM accuracy + speed
• Query Score: Queries served / Total queries
• Uptime Multiplier: (uptime_percentage)² - rewards reliability

Token Distribution ($REM)

Total Supply: 1 Billion REM

• Mining Rewards: 55% (550M REM)
• Team: 10% (100M REM)
• Referrals: 10% (100M REM)
• Marketing: 10% (100M REM)
• CEX/DEX Listings: 10% (100M REM)
• Treasury: 5% (50M REM)

Product Layer & Developer API

REM Network provides a Pinecone-compatible REST API and Python SDK for building AI applications on decentralized infrastructure.

API Endpoints

Base URL: https://api.getrem.online/v1

• POST /collections - Create a collection
• POST /collections/{id}/vectors/upsert - Insert/update vectors
• POST /collections/{id}/vectors/query - Similarity search
• POST /collections/{id}/vectors/fetch - Fetch vectors by ID
• POST /collections/{id}/vectors/delete - Delete vectors
• POST /collections/{id}/vectors/query/batch - Batch queries (up to 10)

Python SDK

pip install rem-vectordb

from rem import REM

client = REM(api_key="rem_your_api_key")

# Create a collection with encryption
collection = client.create_collection(
    name="my-docs",
    dimension=384,
    metric="cosine",
    encrypted_fields=["text", "user_id"]
)

# Upsert vectors
collection.upsert(vectors=[
    {"id": "doc-1", "values": [0.1, 0.2, ...], "metadata": {"text": "Hello world"}}
])

# Query with metadata filtering
results = collection.query(
    vector=[0.1, 0.2, ...],
    top_k=10,
    filter={"category": {"$eq": "science"}},
    include_metadata=True
)

# Hybrid search (vector + keyword)
results = collection.query(
    vector=[0.1, 0.2, ...],
    query_text="machine learning",
    hybrid_alpha=0.7,
    top_k=10
)

# Fetch and delete
vectors = collection.fetch(ids=["doc-1"])
collection.delete(ids=["doc-1"])

Key Features

• AES-256-GCM Encryption - Per-field metadata encryption with per-namespace keys
• Vector Obfuscation - Distance-preserving permutation + sign flip before reaching miners
• Hybrid Search - Combine vector similarity with BM25 keyword search via Reciprocal Rank Fusion
• Metadata Filtering - Pinecone-compatible operators: $eq, $gt, $gte, $lt, $lte, $in, $nin, $and, $or
• Batch Queries - Execute up to 10 queries in a single API call
• Idempotency - Upsert and delete support Idempotency-Key header (24h TTL)
• LangChain & LlamaIndex - Native integrations for RAG pipelines

Framework Integrations

# LangChain
from rem.integrations.langchain import REMVectorStore
store = REMVectorStore(api_key="rem_xxx", collection_name="docs", embedding=embeddings)
store.add_texts(["Hello world", "REM is great"])
results = store.similarity_search("greeting", k=5)

# LlamaIndex
from rem.integrations.llamaindex import REMVectorStore
vector_store = REMVectorStore(api_key="rem_xxx", collection_name="docs")
index = VectorStoreIndex.from_vector_store(vector_store)

Network Status

Current Configuration:

• Epoch Duration: 1 hour
• Network: Sui Testnet
• Replication Factor: 3×
• Status: Testnet Active

How to Mine

Ready to start contributing? Check out our Mining Guide for step-by-step instructions on downloading and running the miner package.

Community & Support

• Discord: discord.gg/9ndMQY4PYP
• X (Twitter): @RemNetwork
• Telegram: t.me/RemDepin
• Explorer: Network Stats