Dynamic trees, made fast.

HiraOS is a tree engine built around a novel tree algorithm developed by Jaun van Heerden. This demo lets you benchmark it on realistic workloads, sweep complexity curves, and upload your own trees. The structure powering it is described at a high level in the linked whitepaper; the internals remain proprietary.

A note on the demo's size limit

To keep this public demo responsive and its memory footprint predictable, the engine here runs in a fixed-width label mode. Node identifiers are bounded, which caps how deep or wide a single tree can grow before you'll see a label limit reached message. That cap is a property of this build, not the algorithm.

The algorithm itself has no inherent size limit and, in its unbounded form, scales to arbitrarily large trees (billions of nodes, arbitrary depth) with the same complexity profile shown on the benchmark and complexity pages. The unbounded form is the same algorithm, just with identifiers sized to the workload instead of pre-clipped for an interactive sandbox. If you want to evaluate it at production scale, reach out.

What are dynamic trees?

A dynamic tree is any rooted tree that changes shape at runtime. Nodes are inserted, removed, and moved between parents, and queries (ancestor checks, subtree reads, child lookups) interleave with those edits. They show up wherever hierarchy is not static: org charts, file systems, scene graphs, permission and policy stores, dependency graphs, taxonomies, multi-tenant resource trees, and BOM/asset hierarchies.

The hard part is that the cheap representations (parent pointers, adjacency lists) make some operations free and others linear or worse, and the operations you actually need depend on the workload. Established dynamic-tree algorithms trade constant factors and code complexity for asymptotic guarantees, and most are tuned for graph-theoretic problems rather than the mixed read/write patterns of real applications.

Why this is different

The algorithm behind HiraOS answers the questions that dominate real workloads (is X an ancestor of Y, what is the depth of N, where does this subtree live) without the traversal cost classical structures pay for them. Edits stay close to constant-time amortized in the common case, with predictable worst-case behaviour for the rare expensive event.

Constant-time ancestry & depth. Reads that classical structures pay log-depth for collapse to near-trivial work.
Cache-friendly addressing. Flat, allocator-managed identifiers, with no pointer chasing or balanced-tree rebalancing in the hot path.
Workload-tunable. A small number of build-time parameters lets the same engine bias toward read-heavy or write-heavy regimes without changing algorithm.
Predictable tails. The occasional expensive operation is bounded and rare, making p99 latency tractable in ways many established approaches struggle with.

This demo intentionally exposes only the API surface and benchmark results. For a deeper discussion of the approach and trade-offs, see the whitepaper.

Where it pays off

Authorization & multi-tenant hierarchies. Cloud platforms model accounts, orgs, projects, and resources as deep trees and check ancestry on every request. This algorithm turns that check into near-trivial work.
Content & document trees. CMS, knowledge bases, and collaborative editors need to reparent and reorder large subtrees live without locking the whole hierarchy.
Engineering & industrial BOMs. Bills of materials, plant hierarchies, and digital-twin structures evolve constantly and demand auditable, fast subtree queries.
Game & simulation scene graphs. Entity hierarchies see thousands of mutations per frame; the engine keeps per-op cost low enough for real-time loops.
Taxonomies, catalogs, and search. Retail categories, biological taxonomies, and faceted indexes that update without offline rebuilds.

For the formal treatment and benchmarks against established dynamic-tree algorithms, see the whitepaper.

Terms & notices

© 2026 Jaun van Heerden. All rights reserved. HiraOS™, the HiraOS logo, and associated branding are trademarks of Jaun van Heerden. The underlying algorithm, its implementation, the API surface, the benchmark methodology, and all related know-how are the proprietary and confidential property of Jaun van Heerden and are protected as trade secrets and by applicable copyright, trademark, and other intellectual-property laws.

No licence is granted. Access to this demo is permitted solely for individual, non-commercial evaluation. No right or licence, whether express, implied, by estoppel, or otherwise, is granted to any patent, copyright, trademark, trade secret, or other intellectual-property right of Jaun van Heerden, except as expressly stated in writing and signed by the author.

Prohibited conduct. You may not, and may not permit any third party to: (a) reverse-engineer, decompile, disassemble, observe the behaviour of, probe, profile, fuzz, scrape, or otherwise attempt to discover or derive the algorithm, source code, internal data structures, parameters, or implementation details of the service; (b) use, copy, reproduce, modify, translate, adapt, create derivative works of, redistribute, sublicense, sell, rent, lease, or otherwise commercially exploit any part of the service, its outputs, the whitepaper, or any artefact produced by or derived from it; (c) republish, quote, or reference benchmark results, screenshots, or other output without the author's prior written consent; (d) use the service or any output to develop, train, or evaluate a competing product, model, algorithm, or service, including any machine-learning model; (e) remove, obscure, or alter any proprietary notices, trademarks, or attribution; (f) use the service to violate any law, infringe any right, or for any purpose other than the evaluation permitted above; or (g) circumvent or attempt to circumvent any rate limits, access controls, or other technical measures.

Confidentiality. Any information you obtain through the service that is not expressly published (including engine internals, error semantics, parameter ranges, latency profiles, and any communications with the author marked or reasonably understood as confidential) is the confidential information of Jaun van Heerden. You agree to hold such information in strict confidence, to use it solely for evaluation, and not to disclose it to any third party. Technical detail beyond what is publicly documented is shared only under a mutual non-disclosure agreement.

No warranty. The service is provided “as is” and “as available”, without warranty of any kind, whether express, implied, statutory, or otherwise, including all warranties of merchantability, fitness for a particular purpose, title, non-infringement, accuracy, reliability, and quiet enjoyment. Benchmark figures reflect a specific environment, configuration, and workload and are not representations or guarantees of production performance.

Limitation of liability. To the maximum extent permitted by law, Jaun van Heerden shall not be liable for any indirect, incidental, special, consequential, exemplary, or punitive damages, or for any loss of profits, revenue, data, goodwill, or business opportunities, arising out of or in connection with your use of, or inability to use, the service, even if advised of the possibility of such damages. Total aggregate liability shall not exceed five United States dollars (USD 5).

Indemnity. You agree to indemnify, defend, and hold harmless Jaun van Heerden from and against any claims, liabilities, damages, losses, and expenses (including reasonable legal fees) arising out of or in any way connected with your breach of these terms or your misuse of the service or its outputs.

Governing law. These terms are governed by the laws of the jurisdiction of the author's primary residence, without regard to conflict-of-laws principles. The exclusive forum for any dispute arising under or relating to these terms shall be the competent courts of that jurisdiction, and you consent to personal jurisdiction therein.

Reservation of rights. All rights not expressly granted are reserved by Jaun van Heerden. No failure or delay in enforcing any term constitutes a waiver. If any provision is held unenforceable, the remainder shall remain in full force. These terms may be updated at any time; continued use constitutes acceptance.