Semantic AI Query Language

SAIQL Technical Overview

SAIQL is a semantic query language and database engine built for AI systems. It replaces verbose SQL + JSON stacks with symbolic queries and LoreToken compression, giving models a compact, unambiguous way to read and write structured knowledge. High-level explanation: saiql.ai.

1. Repositories

Core engine and reference implementation live in:

apolloraines/SAIQL-Engine_v0.2.1 – engine, server, benchmarks.
apolloraines/LoreTokens – LoreToken format kit and compression tools.

Use the site pages (this doc + loretokens.ai) for narrative explanations, and GitHub for full technical details.

saiql engine loretokens semantic compression ai database nova live use

2. SAIQL Language Basics

A SAIQL query is short, regular, and explicit enough that the model does not have to guess what it means. It encodes target set, filters, projections, and output shape in one compact line.

Select a few users

*3[users]::id,name,email>>flat

Means: “Give me three rows from users, with only id, name, and email, in a flat list shape.”

Filter with constraints

*[orders?(total>=100)&(status:="paid")]::id,total,created_at>>time_series

Filtering is expressed with symbolic predicates. The engine converts this into an execution plan over indexes and compressed storage.

For runnable examples, see:

demo_saiql.py – end-to-end demo script.
examples/ – assorted query patterns.

SAIQL is deliberately not “pretty SQL”. It is tuned for token efficiency and pattern recognition in LLMs, while remaining readable to engineers.

3. Engine & Performance

The SAIQL Engine is a compact database core that uses symbol-aware indexes and execution plans to achieve far higher throughput than a generic SQL setup.

Install

Quickstart & Scripts

Start with QUICKSTART.md , then use install_minimal.sh or install_system.sh .

Benchmarks

220× vs PostgreSQL

Benchmark overview is in benchmark.md , with scripts in benchmarks/ and the speed_test/ harness.

Positioning

SAIQL vs AI Databases

How SAIQL compares to vector DBs and SQL: SAIQL_vs_AI_Databases.md .

Live Use Example

Nova AI

A complete walkthrough of SAIQL + LoreTokens inside Nova: NOVA_COMPLETE_EXPLANATION.md . This file is also linked from saiql.ai for readers coming from the site.

4. LoreTokens Inside SAIQL

LoreTokens are the semantic compression layer under SAIQL. While SAIQL can address traditional tables, it shines when the underlying data is stored as compact symbolic LoreTokens.

Where to find the LoreToken spec

LoreTokens README – overview & history.
docs/FORMATS.md and docs/SYMBOLS.md – format & glyph reference.
examples/Schema-LoreTokenised.txt – example schemas suitable for LLM training.

SAIQL’s integration tests live in the engine repo:

Token reduction is useful up to the point where the model has to guess meaning. LoreTokens + SAIQL stay on the safe side of that line: fewer tokens than JSON, without turning every query into a guessing game. More examples and use cases are collected at loretokens.ai.

5. Typical Use Cases

AI agents that need fast, repeated access to structured knowledge without blowing up context windows.
Long-term memory layers for systems like Nova, storing rules, histories and configs as LoreTokens.
Hybrid setups where a traditional DB feeds SAIQL, which then exposes a compressed, AI-friendly surface.
Rolling context windows: SAIQL can stream just the right LoreTokens back into the model each step.

6. Key GitHub Entry Points

SAIQL Engine root: SAIQL-Engine_v0.2.1
LoreTokens root: LoreTokens
SAIQL roadmap: ROADMAP.md
LoreTokens use cases: LoreToken_UseCases.md

7. Keyword Index

SAIQL – Semantic AI Query Language – SAIQL Engine – LoreTokens – semantic compression – symbolic database – AI database – AI memory – rolling context – Nova crypto trading AI – JSON replacement – SQL alternative – PostgreSQL performance – compressed knowledge store.