DistX

DistX does not store vectors that represent objects.
It stores objects, and derives vectors from their structure.

A high-performance vector database with the Similarity Contract — a schema-driven approach to structured similarity search that is deterministic, explainable, and requires no external ML.

The Similarity Contract

The schema is not just configuration — it's a contract that governs:

Aspect	What the Schema Controls
Ingest	How objects are converted to vectors (deterministic, reproducible)
Query	How similarity is computed across multiple field types
Ranking	How results are scored with structured distance functions
Explainability	How each field contributes to the final score

This is an architectural difference, not just an API feature. You cannot replicate this with Qdrant hybrid queries without replicating half this codebase client-side.

What DistX Is (and Is Not)

DistX IS	DistX is NOT
A contract-based similarity engine	A neural embedding model
Deterministic and reproducible	A probabilistic LLM system
Designed for structured/tabular data	A black-box recommender
Fully explainable (per-field scores)	Dependent on external ML APIs

Target domains: ERP, e-commerce, CRM, financial data, any tabular dataset.

How It Works

┌──────────────────────────────────────────────────────────────────────────┐
│  Traditional Vector Database                                             │
│  ───────────────────────────                                             │
│  Your Data → External ML API → Embeddings → Vector DB → Score: 0.87     │
│              (cost per call)   (black box)              (unexplained)    │
│              (model drift)     (retraining)             (no breakdown)   │
├──────────────────────────────────────────────────────────────────────────┤
│  DistX with Similarity Contract                                          │
│  ──────────────────────────────                                          │
│  Your Data → Schema (JSON) → Deterministic → Explainable Results        │
│              (contract)       (no drift)     (name: 0.25, price: 0.22)   │
│              (stable)         (reproducible) (auditable)                 │
└──────────────────────────────────────────────────────────────────────────┘

📖 Detailed comparison with Qdrant, Pinecone, Elasticsearch →

Similarity Contract Engine

The first schema-driven structured similarity engine with built-in explainability.

Define a Similarity Contract, insert your data, and query by example — vectors are derived automatically from object structure. No external ML, no embedding pipelines, no black-box scores.

# 1. Define similarity schema
curl -X PUT http://localhost:6333/collections/products -H "Content-Type: application/json" -d '{
  "similarity_schema": {
    "fields": {
      "name": {"type": "text", "weight": 0.4},
      "price": {"type": "number", "distance": "relative", "weight": 0.3},
      "category": {"type": "categorical", "weight": 0.2},
      "brand": {"type": "categorical", "weight": 0.1}
    }
  }
}'

# 2. Insert data (vectors auto-generated)
curl -X PUT http://localhost:6333/collections/products/points -H "Content-Type: application/json" -d '{
  "points": [
    {"id": 1, "payload": {"name": "Prosciutto di Parma DOP", "price": 8.99, "category": "salumi", "brand": "Parma"}},
    {"id": 2, "payload": {"name": "Prosciutto cotto", "price": 4.99, "category": "salumi", "brand": "Negroni"}},
    {"id": 3, "payload": {"name": "iPhone 15 Pro", "price": 1199, "category": "electronics", "brand": "Apple"}}
  ]
}'

# 3. Query by example
curl -X POST http://localhost:6333/collections/products/similar -H "Content-Type: application/json" \
  -d '{"example": {"name": "prosciutto crudo", "price": 8.0, "category": "salumi"}, "limit": 3}'

Response includes per-field explainability:

┌──────┬─────────────────────────────┬───────┬──────────────────────────────────┐
│ Rank │ Product                     │ Score │ Contribution Breakdown           │
├──────┼─────────────────────────────┼───────┼──────────────────────────────────┤
│  1   │ Prosciutto di Parma DOP     │ 0.71  │ name: 0.22, price: 0.22          │
│  2   │ Prosciutto cotto            │ 0.68  │ name: 0.25, category: 0.20       │
│  3   │ Coppa di Parma              │ 0.53  │ category: 0.20, price: 0.25      │
└──────┴─────────────────────────────┴───────┴──────────────────────────────────┘

Key Capabilities

Capability	Description
Schema-Driven	Declarative field definitions with typed similarity (text, number, categorical, boolean)
Auto-Embedding	Deterministic vector generation from structured payloads
Query by Example	Natural JSON queries instead of raw vectors
Explainable Scoring	Per-field contribution breakdown for every result
Dynamic Weights	Override field importance at query time without re-indexing
Zero External Dependencies	Fully self-contained, works offline and air-gapped

What You Can Do That Qdrant Cannot

Example 1: Change Similarity Semantics Without Re-embedding

# Same data, different meaning of "similar" — no re-indexing required

# Query 1: "Find similar products" (balanced)
curl -X POST /collections/products/similar \
  -d '{"example": {"name": "iPhone 15"}, "limit": 5}'

# Query 2: "Find cheaper alternatives" (boost price)
curl -X POST /collections/products/similar \
  -d '{"example": {"name": "iPhone 15"}, "weights": {"price": 0.7, "name": 0.2}, "limit": 5}'

# Query 3: "Find same brand, any price" (boost brand)
curl -X POST /collections/products/similar \
  -d '{"example": {"name": "iPhone 15"}, "weights": {"brand": 0.6, "category": 0.3}, "limit": 5}'

In Qdrant: You would need to re-embed everything or build complex client-side logic.

Example 2: Same Schema, Different Datasets

# One Similarity Contract works across domains:

# Products
{"name": "iPhone 15", "price": 999, "category": "electronics", "brand": "Apple"}

# Suppliers  
{"name": "Acme Corp", "price": 5000000, "category": "manufacturing", "brand": "certified"}

# Financial Assets
{"name": "AAPL Stock", "price": 178.50, "category": "equity", "brand": "tech"}

# Same schema, same queries, same explainability — across all datasets

This is not product-specific. The Similarity Contract is domain-agnostic.

📖 Documentation · Interactive Demo · Comparison with Alternatives

100% Qdrant API Compatible

DistX maintains full compatibility with the Qdrant API, so you can:

✅ Use existing Qdrant client libraries (Python, JavaScript, Rust, Go)
✅ Drop-in replace Qdrant in your stack
✅ Use Qdrant's Web Dashboard UI
✅ Migrate with zero code changes

The Similarity Engine is additive — all standard vector operations work exactly like Qdrant:

# Standard Qdrant-compatible vector search still works!
curl -X POST http://localhost:6333/collections/my_collection/points/search \
  -d '{"vector": [0.1, 0.2, 0.3, ...], "limit": 10}'

Quick Start

1. Start DistX with Docker

# Pull and run (with persistent storage)
docker run -d --name distx \
  -p 6333:6333 -p 6334:6334 \
  -v distx_data:/qdrant/storage \
  distx/distx:latest

# Or with docker-compose
docker-compose up -d

DistX is now running at:

REST API: http://localhost:6333
Web Dashboard: http://localhost:6333/dashboard
gRPC: localhost:6334

2. Create a Collection with Similarity Schema

curl -X PUT http://localhost:6333/collections/products \
  -H "Content-Type: application/json" \
  -d '{
    "similarity_schema": {
      "fields": {
        "name": {"type": "text", "weight": 0.4},
        "price": {"type": "number", "distance": "relative", "weight": 0.3},
        "category": {"type": "categorical", "weight": 0.2},
        "in_stock": {"type": "boolean", "weight": 0.1}
      }
    }
  }'

3. Insert Data (No Vectors Needed!)

curl -X PUT http://localhost:6333/collections/products/points \
  -H "Content-Type: application/json" \
  -d '{
    "points": [
      {"id": 1, "payload": {"name": "Prosciutto di Parma DOP", "price": 8.99, "category": "salumi", "in_stock": true}},
      {"id": 2, "payload": {"name": "Prosciutto cotto", "price": 4.99, "category": "salumi", "in_stock": true}},
      {"id": 3, "payload": {"name": "Mortadella Bologna", "price": 3.99, "category": "salumi", "in_stock": false}},
      {"id": 4, "payload": {"name": "Parmigiano Reggiano", "price": 18.99, "category": "cheese", "in_stock": true}},
      {"id": 5, "payload": {"name": "Grana Padano", "price": 14.99, "category": "cheese", "in_stock": true}}
    ]
  }'

4. Query by Example

# Find products similar to "prosciutto crudo around $8"
curl -X POST http://localhost:6333/collections/products/similar \
  -H "Content-Type: application/json" \
  -d '{
    "example": {
      "name": "prosciutto crudo",
      "price": 8.0,
      "category": "salumi"
    },
    "limit": 3
  }'

Response with explainable scores:

{
  "result": [
    {
      "id": 1,
      "score": 0.71,
      "payload": {"name": "Prosciutto di Parma DOP", "price": 8.99, "category": "salumi"},
      "explain": {"name": 0.22, "price": 0.24, "category": 0.20, "in_stock": 0.05}
    },
    {
      "id": 2,
      "score": 0.65,
      "payload": {"name": "Prosciutto cotto", "price": 4.99, "category": "salumi"},
      "explain": {"name": 0.25, "price": 0.15, "category": 0.20, "in_stock": 0.05}
    }
  ]
}

5. Dynamic Weight Overrides

# Find cheaper alternatives (boost price importance)
curl -X POST http://localhost:6333/collections/products/similar \
  -H "Content-Type: application/json" \
  -d '{
    "example": {"name": "prosciutto", "category": "salumi"},
    "weights": {"price": 0.6, "name": 0.2, "category": 0.2},
    "limit": 3
  }'

6. Query by Existing Point ID

# Find products similar to ID 4 (Parmigiano Reggiano)
curl -X POST http://localhost:6333/collections/products/similar \
  -H "Content-Type: application/json" \
  -d '{"like_id": 4, "limit": 3}'

7. Run the Interactive Demo

# Full demo with sample data
python scripts/similarity_demo.py

# Or run specific demos
python scripts/similarity_demo.py --demo products
python scripts/similarity_demo.py --demo suppliers

Alternative: Traditional Vector Search

DistX also supports standard Qdrant-compatible vector operations:

# Create collection with vectors
curl -X PUT http://localhost:6333/collections/embeddings \
  -H "Content-Type: application/json" \
  -d '{"vectors": {"size": 128, "distance": "Cosine"}}'

# Insert with vectors
curl -X PUT http://localhost:6333/collections/embeddings/points \
  -H "Content-Type: application/json" \
  -d '{
    "points": [
      {"id": 1, "vector": [0.1, 0.2, ...], "payload": {"text": "example"}}
    ]
  }'

# Vector search
curl -X POST http://localhost:6333/collections/embeddings/points/search \
  -H "Content-Type: application/json" \
  -d '{"vector": [0.1, 0.2, ...], "limit": 10}'

Installation Alternatives

# From crates.io
cargo install distx
distx --data-dir ./data

# From source
git clone https://github.com/antonellof/distx
cd distx && cargo build --release
./target/release/distx

Performance

Metric	Performance
Vector Insert	~8,000 ops/sec
Vector Search	~400-500 ops/sec
Search Latency (p50)	~2ms
Search Latency (p99)	~5ms
Similarity Query	<1ms overhead

Benchmarks: 5,000 vectors, 128 dimensions, Cosine distance

All Features

Similarity Engine (NEW)

Schema-driven similarity — Define what fields matter
Auto-embedding — Vectors generated from payload
Multi-type support — Text, number, categorical, boolean
Explainable results — Per-field score breakdown
Dynamic weights — Override at query time

Vector Database

HNSW Index — Fast ANN with SIMD (AVX2, SSE, NEON)
BM25 Text Search — Full-text ranking
Payload Filtering — JSON metadata queries
Dual API — REST + gRPC
Persistence — WAL, snapshots, LMDB

Operations

Single Binary — ~6MB, no dependencies
Docker Ready — Single command deployment
Web Dashboard — Qdrant-compatible UI

Documentation

Guide	Description
Similarity Engine	Schema-driven similarity for tabular data
Similarity Demo	Interactive walkthrough with examples
Comparison	DistX vs Qdrant, Pinecone, Elasticsearch
Quick Start	Get started in 5 minutes
Docker Guide	Container deployment
API Reference	REST and gRPC endpoints
Architecture	System design

Use Cases

🛒 E-Commerce & Retail

Problem: "Show me products similar to this one" — but similarity means different things (style, price, brand).

# Similar products for "customers also viewed"
curl -X POST /collections/products/similar -d '{
  "example": {"name": "Nike Air Max 90", "price": 129, "category": "sneakers", "brand": "Nike"},
  "limit": 6
}'

# Budget alternatives (boost price importance)
curl -X POST /collections/products/similar -d '{
  "like_id": 123,
  "weights": {"price": 0.6, "category": 0.3, "brand": 0.1}
}'

Use cases:

"Similar products" on product pages
"You might also like" recommendations
Competitor price matching (find similar products, compare prices)
Inventory substitution (out of stock → suggest alternatives)

🏭 ERP & Supply Chain

Problem: Find the best supplier match based on multiple criteria without building ML pipelines.

# Find suppliers similar to your top performer
curl -X POST /collections/suppliers/similar -d '{
  "example": {
    "industry": "manufacturing",
    "annual_revenue": 5000000,
    "employee_count": 150,
    "certified": true,
    "location": "Milan"
  },
  "limit": 10
}'

Use cases:

Supplier discovery and matching
Vendor risk assessment (find similar vendors to flagged ones)
Partner recommendations
RFQ (Request for Quote) matching

👥 CRM & Customer Data

Problem: Find similar customers for segmentation, lead scoring, or churn prediction.

# Find customers similar to your best ones
curl -X POST /collections/customers/similar -d '{
  "example": {
    "industry": "fintech",
    "company_size": "enterprise",
    "annual_spend": 50000,
    "engagement_score": 85
  }
}'

# Find leads similar to closed-won deals
curl -X POST /collections/leads/similar -d '{
  "like_id": "deal_12345",
  "weights": {"deal_size": 0.4, "industry": 0.3, "company_size": 0.3}
}'

Use cases:

Lead scoring (similar to converted leads?)
Customer segmentation
Churn prediction (similar to churned customers?)
Account-based marketing (find lookalike companies)

🔍 Data Quality & Deduplication

Problem: Find duplicate or near-duplicate records without exact matching.

# Find potential duplicates
curl -X POST /collections/contacts/similar -d '{
  "example": {"name": "John Smith", "email": "j.smith@acme.com", "company": "Acme Inc"},
  "limit": 5
}'

# Response shows WHY records might be duplicates
# → name: 0.35 (similar names)
# → company: 0.25 (same company) 
# → email: 0.15 (different email domain)

Use cases:

Contact/account deduplication
Data cleansing before migration
Master data management
Merge candidate identification

📊 Data Analysis & Exploration

Problem: Explore datasets by finding similar records without writing complex SQL.

# "Find transactions similar to this suspicious one"
curl -X POST /collections/transactions/similar -d '{
  "example": {"amount": 9999, "merchant_category": "travel", "country": "unusual"},
  "weights": {"amount": 0.5, "merchant_category": 0.3}
}'

# "Find properties similar to this sold one"
curl -X POST /collections/properties/similar -d '{
  "example": {"sqft": 2500, "bedrooms": 4, "neighborhood": "downtown", "year_built": 2010}
}'

Use cases:

Fraud pattern detection
Anomaly investigation
Comparable analysis (real estate, finance)
Research dataset exploration

⚖️ Regulated Industries (Finance, Healthcare, Legal)

Problem: Need similarity search with full auditability — can't use black-box ML.

Why DistX:

Explainable scores — Per-field contribution breakdown
Deterministic — Same query always returns same explanation
Auditable — Schema defines what matters, weights are transparent
No external APIs — Data never leaves your infrastructure

# Healthcare: Find similar patient cases
curl -X POST /collections/patients/similar -d '{
  "example": {"diagnosis_code": "E11.9", "age_group": "65+", "comorbidities": 3}
}'

# Response includes full explanation for audit trail:
# {
#   "score": 0.78,
#   "explain": {
#     "diagnosis_code": 0.35,  ← Same diagnosis
#     "age_group": 0.25,       ← Same age bracket
#     "comorbidities": 0.18   ← Similar complexity
#   }
# }

Use cases:

Clinical trial patient matching
Insurance claim similarity
Legal case precedent search
Compliance reporting

🏠 Real Estate & Property

# Find comparable properties for valuation
curl -X POST /collections/properties/similar -d '{
  "example": {
    "sqft": 2200,
    "bedrooms": 3,
    "bathrooms": 2,
    "year_built": 2015,
    "neighborhood": "downtown",
    "property_type": "condo"
  },
  "weights": {"sqft": 0.3, "neighborhood": 0.25, "property_type": 0.2}
}'

Use cases:

Comparable property analysis (comps)
Property valuation
Investment opportunity matching
Tenant-property matching

🎯 HR & Recruiting

# Find candidates similar to your top performers
curl -X POST /collections/employees/similar -d '{
  "example": {
    "department": "engineering",
    "years_experience": 5,
    "skills": "rust,python",
    "performance_rating": "exceeds"
  }
}'

Use cases:

Candidate matching to job requirements
Internal mobility (find similar roles)
Team composition analysis
Succession planning

Use as a Library

[dependencies]
distx = "0.2.5"
distx-schema = "0.2.5"  # Similarity Engine
distx-core = "0.2.5"        # Core data structures

use distx_similarity::{SimilaritySchema, FieldConfig, StructuredEmbedder, Reranker};
use std::collections::HashMap;

// Define schema
let mut fields = HashMap::new();
fields.insert("name".to_string(), FieldConfig::text(0.5));
fields.insert("price".to_string(), FieldConfig::number(0.3, DistanceType::Relative));
fields.insert("category".to_string(), FieldConfig::categorical(0.2));

let schema = SimilaritySchema::new(fields);
let embedder = StructuredEmbedder::new(schema.clone());

// Auto-generate vector from payload
let payload = json!({"name": "Prosciutto", "price": 8.99, "category": "salumi"});
let vector = embedder.embed(&payload);

License

Licensed under MIT OR Apache-2.0 at your option.

distx-schema 0.2.7

DistX

The Similarity Contract

What DistX Is (and Is Not)

How It Works

Similarity Contract Engine

Key Capabilities

What You Can Do That Qdrant Cannot

Example 1: Change Similarity Semantics Without Re-embedding

Example 2: Same Schema, Different Datasets

100% Qdrant API Compatible

Quick Start

1. Start DistX with Docker

2. Create a Collection with Similarity Schema

3. Insert Data (No Vectors Needed!)

4. Query by Example

5. Dynamic Weight Overrides

6. Query by Existing Point ID

7. Run the Interactive Demo

Alternative: Traditional Vector Search

Installation Alternatives

Performance

All Features

Similarity Engine (NEW)

Vector Database

Operations

Documentation

Use Cases

🛒 E-Commerce & Retail

🏭 ERP & Supply Chain

👥 CRM & Customer Data

🔍 Data Quality & Deduplication

📊 Data Analysis & Exploration

⚖️ Regulated Industries (Finance, Healthcare, Legal)

🏠 Real Estate & Property

🎯 HR & Recruiting

Use as a Library

Links

License