SAIQL
Semantic AI Query Language · Pronounced "Saykwill"
An AI-native database and query language purpose-built for AI systems — not a wrapper around a vector store, not a retrofitted RDBMS. SAIQL unifies semantic search, structured retrieval, agent memory, and deterministic RAG under a single symbolic query interface. It also handles enterprise database migration when you need to move data between systems.
ATLAS — Lock-Weight RAG (LRAG)
Standard RAG was designed in 2020 for academic NLP benchmarks — research datasets where "close enough most of the time" is acceptable. It was then cargo-culted into medical, legal, and financial production systems without re-examining the foundational assumptions. Those assumptions don't hold. Approximate nearest neighbor search is approximate by definition: same query, different results, every time. ATLAS was built to solve that.
The First Deterministic RAG System
ATLAS uses three fused retrieval lanes — metadata filtering, lexical search, and semantic similarity — combined with locked fusion weights, content-addressable chunk IDs (SHA-256), and lexicographic tie-breaking. The result: the same query returns the same results every time, without exception. ATLAS CE is open source and published on GitHub.
Architecture
QIPI Storage Engine
Quantum-Inspired Probabilistic Index. Three-layer architecture: L1 hot cache, L2 bloom filter, L3 hash buckets. 10x+ faster lookups than traditional B-tree indexes for AI query patterns. Built in — no external database required.
LoreCore
LSM-tree agent-native storage with event streams and deterministic clock. Designed for persistent AI agent memory — modules, decisions, event history, and identity survive sessions, migrations, and restarts.
LoreTokens
Semantic compression architecture with three modes: parametric pointer compression (model as decompressor, unbounded ratio), structural data compression (up to 2.6x), and temporal context compression (eliminates repeated re-transmission across calls).
Semantic Firewall
Pre-query, pre-retrieval, and post-output security guards. Protects against prompt injection, malicious query construction, and unsafe output. Fails closed — if it can't verify safety, it doesn't proceed.
Multi-Tenant Architecture
Role-based access control with three tiers (Viewer / Operator / Admin). Tenant isolation at the storage layer. Profile-based credentials — secrets never inline, never logged.
Universal Query Transpiler
Translates SAIQL symbolic queries across any database dialect via an intermediate representation (IR) layer. Bridges AI systems to legacy databases without requiring schema changes or database-specific integrations.
Enterprise Database Migration
Moving data between database systems is an unavoidable part of AI infrastructure work. SAIQL handles it natively — schema introspection, type-safe conversion through the IR layer, proof-first validation, and immutable artifact bundles for audit compliance. Supports 8+ database engines.
Introspection
Connect to source and extract complete schema: tables, columns, types, constraints, indexes, procedures.
Type Mapping
Convert types through IR layer (Source → IR → Target). Lossy conversions — precision loss, timezone stripping — are flagged explicitly.
Schema Generation
Generate target DDL with CREATE TABLE, constraints, indexes, and defaults. Dry-run mode before any changes are made.
Data Extraction
Deterministic PK-ordered extraction with batching, checkpoints, and recovery. Output to CSV, Parquet, or direct DB load.
Validation
Row counts, checksums, constraint parity. Audit report with explicit warnings for lossy conversions or manual steps required.
Artifact Bundle
Immutable bundle: DDL, data files, validation reports, limitations docs, and execution logs. Compliance-grade output.
| Database | Source | Target | Level | Status |
|---|---|---|---|---|
| PostgreSQL | ✅ | ✅ | L1 | Production |
| MySQL | ✅ | ✅ | L1 | Production |
| SQLite | ✅ | ✅ | L1 | Production |
| MariaDB | ✅ | ✅ | L1 | Production |
| SAP HANA | ✅ | — | L0/L1 | Beta |
| Oracle | ✅ | — | L0/L1 | Beta |
| SQL Server | ✅ | ✅ | L1 | Beta |
| DuckDB | ✅ | ✅ | L1 | Experimental |
Proven In Production
SAIQL is not a demo. It is the infrastructure layer powering two live products today.
Architecture Papers
The SAIQL architecture is documented in three peer-reviewed preprints published on Zenodo with permanent DOIs. The math is published. We are not asking you to trust us.
Investor or Early Access Inquiries
SAIQL is available to select partners and investors. Contact Apollo Raines to discuss the technology, roadmap, or early access.