Release Jan 15, 2026 Updated Oct 14, 2026 10 min read

Introducing HeliosDB Full v7.1.0

Editor's note — October 2026. This post announced v7.1.0 in January 2026, and several of its statements do not describe the current release. HeliosDB Full serves the PostgreSQL, MySQL, Oracle and Redis wire protocols plus an HTTP gateway; MongoDB, Cassandra and SQL Server are Planned in the status summary of the Release and Capability Matrix, which states the best result on any version. There is no listener for DB2, Snowflake, Databricks, Pinecone, ClickHouse or S3. HeliosDB Full 8.3.0 runs as a single server; Raft clustering and multi-region replication are planned. GPU acceleration (NVIDIA CUDA) is available in HeliosDB Full builds on request and is not enabled in the public 8.3.0 release build; AMD ROCm support is planned. In-database ML training, the neural planner, NL-to-SQL and quantum-inspired optimization are planned. HeliosDB Full is not published on crates.io; licensed customers build it from the source distribution. The cluster and GPU benchmark figures first published here have been removed because the 8.3.0 release build cannot reproduce them.

Today we are thrilled to announce the release of HeliosDB Full v7.1.0 — our most ambitious release to date. This version represents a major leap forward in what a single database can do, bringing 193 Rust crates, 14 native protocol adapters, quantum-inspired optimization, and GPU acceleration into a unified platform.

What's New in v7.1.0

HeliosDB Full v7.1.0 is the culmination of months of engineering effort across our entire 193-crate workspace. Here are the highlights:

14 Native Protocol Adapters — Connect using PostgreSQL, MySQL, MongoDB, Redis, Cassandra, Oracle, SQL Server, DB2, Snowflake, Databricks, Pinecone, ClickHouse, HTTP, and S3 wire protocols. One database, every interface your application needs. Each adapter implements the native wire protocol, so existing tools, ORMs, and drivers work without modification.

Quantum-Inspired Query Optimizer — Our Tier 1 safe integration of the quantum optimizer uses simulated quantum annealing for complex join ordering. For queries involving 10+ tables, the optimizer explores the search space using 50 annealing iterations (online) or 500 iterations (batch OLAP), finding near-optimal plans that the traditional dynamic programming approach would take exponentially longer to discover. The optimizer includes safety gating: it only activates for queries above the complexity threshold and falls back to the classical optimizer if the quantum-inspired result is worse.

Raft Consensus & Multi-Region Replication — Production-grade distributed consensus with automatic leader election, log replication, and configurable multi-region deployment strategies. Built on the battle-tested raft 0.7 crate with our custom state machine for replicated SQL execution.

GPU Acceleration — CUDA and ROCm support for analytical query execution and vector similarity search.

Architecture Overview

The v7.1.0 release is organized into 193 crates (consolidated from 184 individual crates) spanning storage, query processing, the protocol adapters, clustering, AI/ML, security, multi-tenancy and the API surface.

The server binary is assembled from the workspace with compile-time build features (73 as of 8.3.0), allowing fine-grained control over which subsystems are included in your deployment.

Performance Benchmarks

The benchmark figures first published in this section were measured on a multi-node cluster with GPUs. They are historical and have been removed: the 8.3.0 release build runs as a single server without a GPU path and cannot reproduce them. Figures measured on current releases are on benchmarks.html.

AI/ML Subsystem: 12 Crates

The v7.1.0 AI subsystem is the most comprehensive in any database product:

Cognitive Database Agents — Self-healing database agents that autonomously detect and resolve performance issues, recommend indexes, manage data quality, and tune query plans. The neural planner (3,149 LOC) learns from query execution history to predict optimal plans.

In-Database ML Training — Train ML models directly in SQL without exporting data. Support for 50+ algorithms including gradient boosting, neural networks, and time-series forecasting. The model registry (4,080 LOC) tracks model versions, metrics, and deployment state.

Conversational BI — Natural language queries over your data. Ask "show me revenue by region for Q4" and get SQL + visualizations. Built on our NL2SQL engine with GraphRAG for complex multi-hop queries.

Getting Started

Licensed customers receive HeliosDB Full through sales:

# Container image: provided to licensed customers by HeliosDB sales,
# or built from the Dockerfile in the licensed source.
# The binary is built from the licensed source distribution;
# HeliosDB Full is not published on crates.io.

Compiler Settings

HeliosDB Full compiles with maximum optimization for production deployments:

[profile.release]
lto = true              # Link-time optimization across all crates
codegen-units = 1       # Single codegen unit for maximum optimization
opt-level = 3           # Maximum optimization level

[workspace.lints.clippy]
all = { level = "warn", priority = -1 }

What's Next

We are already working on v7.2.0 with further improvements to our quantum optimization pipeline, enhanced cognitive agents with autonomous schema evolution, and expanded multimodal vector search for image/audio/video retrieval. Stay tuned for updates.

We'd love to hear your feedback. Join us on GitHub or reach out on Discord.

← Back to Blog