SZL HoldingsGoverned AI that can show its work.

Build systems that act under policy, preserve the decision trail, and keep every public claim tied to inspectable evidence.

Decide under policy

a11oy is the governed execution fabric. It evaluates action, policy, measurement, and integrity before a result is admitted.

Operate in the real world

killinchu applies the same decision discipline to counter-UAS inputs and emits a receipt trail for review.

Verify the claim

Source, proof state, receipts, runtime status, and current limitations live beside the product story—not in a separate trust theater.

One company. Three ways in.

InvestorSee the category and wedgeUnderstand the platform, product surfaces, governance moat, and what is shipping versus roadmap.DeveloperGet to a working callChoose an API or SDK, run the quickstart, inspect the response contract, and verify a receipt.EvaluatorAudit before you trustStart with evidence, proof state, lineage, runtime status, and explicit limitations.

The system today

REAL

a11oy execution fabric

Governed command and action infrastructure with public source, API documentation, evidence surfaces, and explicit runtime status.

Platform dossier →

REAL

killinchu counter-UAS

A domain product that turns airspace inputs into bounded decisions and reviewable receipts. Product claims remain linked to source and evidence.

Product dossier →

ROADMAP

Frontier roles

Provenance Anchor and Policy remain roadmap roles; Operator is documented as a frontier role. These labels are not upgraded by presentation.

Status map →

What makes the architecture different

Most AI stacks optimize the answer. SZL focuses on the act: what was requested, what policy admitted, which evidence was available, what was unavailable, and which receipt lets another party replay the decision boundary.

01 · Bound

Define the permitted action space.
Policy and resource constraints narrow what the system may do before execution.

02 · Decide

Evaluate the request under measured state.
Unavailable inputs remain unavailable; modeled values stay labeled modeled.

03 · Receipt

Preserve the decision trail.
Source, identity, policy outcome, and evidence references remain available for review.

Read the architecture →

Evidence before adjectives

Question	Start here
What is in source?	GitHub organization
What can a developer run?	Quickstart and Developer hub
What is proven, axiomatized, or still open?	Proof and Doctrine register
What is live versus unavailable?	Runtime status
How do I inspect a receipt?	Verification guide
What is the current supply-chain posture?	Security and compliance

No single page upgrades a claim. REAL, MEASURED, MODELED, ROADMAP, and UNAVAILABLE keep their meaning across product, documentation, GitHub, and evidence surfaces.

Citation

bibtex

@software{szl_holdings_2026,
  author    = {Lutar, Stephen P.},
  title     = {SZL Holdings: Sovereign Governed AI},
  year      = {2026},
  publisher = {Zenodo},
  version   = {Doctrine v11 LOCKED},
  doi       = {10.5281/zenodo.20434276},
  url       = {https://github.com/szl-holdings}
}

This documentation uses the versioned KANCHAY design system. Its generated bundle, source revision, and integrity manifest are committed beside the site theme.