ArchitectureA supervised JVM-class runtime — OLTP on seven engines, OLAP on three. AI-native, MCP-native, observable as plain SQL.Read the architecture
Está viendo la edición Perú. Está viendo la edición Colombia. You're viewing the Pakistan edition. Cambiar a la edición global →Cambiar a la edición global →Switch to the global edition →
IBM TechXchange 2026 · Atlanta

Airtool at IBM TechXchange 2026 — Informix Vector and HNSW.

The Airtool team presents the Informix Vector Type at IBM TechXchange 2026 in Atlanta — a native vector opaque type and HNSW approximate nearest-neighbour index for IBM Informix 14.10, with AVX2 SIMD-optimised distance kernels and full RAG pipeline support.

IBM TechXchange 2026 — Atlanta

A conference presentation — and a complete enterprise implementation.

At TechXchange 2026, the Airtool team presents the complete enterprise implementation of a vector opaque type for IBM Informix 14.10. The presentation covers the full engineering lifecycle: the C data type and its registration and de-registration within the Informix DataBlade API, AVX2 SIMD-optimised distance kernels for cosine, L2, L1 and inner product similarity, SQL operator overloading, and a custom HNSW approximate nearest-neighbour index implemented via the Virtual-Index Interface.

The scope does not stop at the prototype. The presentation addresses memory management within the DataBlade API — the constraints that distinguish Informix extensibility from PostgreSQL's — alongside performance benchmarking, known limitations of the Informix extensibility model relative to pgvector, and production-readiness considerations for Retrieval-Augmented Generation and AI-embedding workloads. The technical paper, available for download below, documents the full implementation and benchmark results head-to-head with PostgreSQL pgvector.

The implementation supports up to 3,072 float32 dimensions, matching the OpenAI text-embedding-3-large model directly. The HNSW index uses hierarchical navigable small-world graphs for sub-linear approximate nearest-neighbour queries, with configurable graph parameters and a storage backend that survives server restarts without additional configuration.

Download the technical paper.

Informix Vector Type — v1.104. Complete implementation documentation: C DataBlade type, AVX2 distance kernels, HNSW index, SQL function reference, pgvector benchmarks, and production-readiness notes for RAG workloads.

Informix Vector Type — a native opaque type via DataBlade.

A C data type that gives Informix a first-class vector column — storage, arithmetic, distance functions and SQL operator overloading. AVX2 SIMD acceleration reduces distance computation time on 2,046-dimension vectors substantially compared to scalar equivalents. The full SQL function reference and installation scripts are in the paper.

HNSW index for approximate nearest-neighbour search on Informix.

Standard B-tree indexes cannot index vectors — the curse of dimensionality makes them ineffective above single-digit dimensions. The HNSW implementation brings approximate nearest-neighbour search to Informix via the Virtual-Index Interface, with configurable graph parameters (M, efConstruction, efSearch) and a storage backend that survives server restarts without extra configuration.

RAG on Informix — embeddings, hybrid queries, AI workloads.

The vector type and HNSW index enable full Retrieval-Augmented Generation pipelines directly on Informix data — hybrid SQL queries combining vector similarity with metadata filters, embedding update patterns, and AI-governed agents that run inference against live Informix records without a secondary vector store. Benchmarked head-to-head against pgvector.

Informix, Db2 and the IBM ecosystem — the full picture.

The vector presentation is the centrepiece, but the Airtool platform covers the full IBM data ecosystem. For Informix, the team offers Tier-1 DBA-grade engineering — partitioning, replication, HA, backup-restore drills, on-call rota — alongside the application layer and AI integration work. For Db2, the platform delivers a modern application surface on top of an existing Db2 data tier: REST and gRPC APIs, microservices, governed AI agents and real-time dashboards operating inside the Db2 permission perimeter, with no rewrite of the data layer. For watsonx deployments, the AI-governance architecture — per-user quotas, permission-scoped inference, audit-grade model accountability — maps directly to enterprise watsonx programmes and the TechXchange 2026 focus on agentic AI.

TechXchange 2026 is the 4th annual edition, held at the Georgia World Congress Centre in Atlanta. Co-located with HashiConf, IDUG North America and the COMMON user group, it is the gathering of the full IBM installed-base technical community — the right audience for this conversation.

Book a meeting at TechXchange 2026.

Georgia World Congress Centre · Atlanta · 26–29 October 2026. An architect will confirm your slot within 24 hours.

We use your details only to schedule the meeting you requested. We do not subscribe you to anything, and we do not pass your details to third parties.

Not attending TechXchange?

The team is available for architecture conversations year-round — remote or in person.