Reference Manual

Colt Language & Architecture Specification

Colt is a strictly-typed, unified data language for defining, querying, and mutating records across relational, key-value, document, and vector databases under provable multi-tenant isolation.

1. Language Overview

Source files use the .colt extension. Colt expressions are compiled into parameterized Abstract Syntax Trees (ASTs) and optimized execution plans by the Colt compiler.

// Declare strict schema and query records
DEFINE COLLECTION users STRICT @backend(neon) {
  id: text @key,
  email: text @unique,
  tier: text @default("free")
}

FIND users SELECT id, email WHERE tier == "pro"

2. Schema Definition

Colt schemas support atomic primitive types: text, int, float, boolean, timestamp, and vector.

DEFINE COLLECTION books STRICT @backend(supabase) {
  id: text @key,
  title: text,
  author: text,
  rating: float @default(4.0),
  embedding: vector(1536)
}

3. Mutations (PUT / DELETE)

Records are inserted or updated using the PUT statement with key targeting:

PUT books:item_101 {
  title: "Database Internals",
  author: "Alex Petrov",
  rating: 4.8
}

4. Querying (FIND)

Query operations support projections, filters, sorting, and pagination:

FIND books SELECT id, title, rating
WHERE rating >= 4.5
ORDER BY rating DESC
LIMIT 20

5. Storage Backends & Routing

ColtaneDB directs queries to specialized backing engines according to the @backend(...) directive:

@backend(redis): Sub-millisecond key-value caching tier on Upstash Redis.
@backend(neon) & @backend(supabase): ACID transactional relational storage.
@backend(firestore): Document collection storage on Google Cloud Firebase.
@backend(turso): Edge-distributed SQLite via libSQL.