Skip to content
@szl-holdings

SZL Holdings

Governed AI infrastructure for inspectable decisions: source, runtime state, receipts, and proof boundaries.

SZL Holdings

Governed intelligence. Checkable results.

SZL Holdings builds AI software, models and research tools that connect an output to its source and evidence. Explore the systems, inspect what is available, and reproduce the work behind each claim.

Explore Command Lab → · Build with the source →

Repeat a public calculation

Input and hash → Executed code → Retained result → Repeat → Limits

The single-cell example summarizes public GEO GSE85241 data and checks integrity and repeatability. Its receipt is unsigned; it makes no biological or clinical claim.

Choose a path

What we build

A11oy connects governed workflows. SZL Router routes model requests through A11oy Code and publishes a status surface. Killinchu demonstrates public observation and operator decisions; public effectors remain SIMULATED. Forge houses model and software research.

A published file, running application and independently checked result establish different things. Each artifact needs its own evidence and use limits.

Current state

The source-bound public inventory records 34 public Spaces, 47 model repositories, 14 native kernels, 37 datasets at 2026-10-04T15:36:53Z under hf-public-author-membership/v2. This is a dated observation, not a live count. Native kernel IDs may also appear in the model namespace; these figures do not count unique projects or trained language models. Inventory binding · Exact source · Current Hub listing. Hub inventory is registry evidence, not availability, operational readiness, or publication policy. It does not establish model quality.

HISTORICAL: estate-alignment v1 described 16 portfolio Spaces, 1 inventory-only Space, 45 models, and 34 datasets. The SZLHOLDINGS/SZLHOLDINGS dataset is a historical profile mirror. It does not establish today's inventory, readiness, or model quality.

Reproduce an example and inspect the architecture

Architecture image · Authorization rules

Portfolio roles

  • Three commercial flagships: A11oy, Killinchu, Forge.
  • One inference flagship: SZL Router.
  • Five public domain bodies: Terra, Killinchu, PRISM Counsel, PURIQ Finance, LYTE.
  • Six internal engines: CHAPAQ (folded into Killinchu), Lyte, Killinchu, Finance, Terra, Counsel.

These are architecture roles. Inspect each artifact's source and stated use limits before choosing a use.

Pinned Loading

  1. a11oy a11oy Public

    a11oy — governed-AI Command Center. Deny by default. Signed receipts. Trust ceiling 0.97. Λ = Conjecture 1.

    Python 1

  2. szl-receipt szl-receipt Public

    Shared signed-receipt library — DSSE/ECDSA-P256, cosign-compatible, UNSIGNED-honest fallback. The one signing primitive every SZL component shares.

    Python

  3. governed-receipt-spec governed-receipt-spec Public

    Open format for a governed AI inference decision receipt (decision, Λ status, measured-or-null energy, hash chain, DSSE) + a dependency-free offline verifier. Receipt-tier trust: honest, replayable…

    Python

  4. lutar-lean lutar-lean Public

    Lean 4 + Mathlib formalization of the Λ aggregator — Λ uniqueness as Conjecture 1 (not a closed theorem). 749 declarations · 14 axioms · 163 tracked sorries. Backs the SZL governance gate. Doctrine…

    Lean

  5. killinchu killinchu Public

    Counter-UAS reference at /elite. Tracks, fusion, ROE, BFT, DSSE per interdiction. Public synthetic — no public effector. Λ = Conjecture 1.

    Python 1

  6. hatun-mcp hatun-mcp Public

    Hatun-MCP — doctrine-aware Model Context Protocol server. 16 SZL tools under PURIQ governance (Yuyay-13 gate, Khipu receipts, DSSE-signed). Streamable HTTP + SSE.

    Python 1

Repositories

Showing 10 of 128 repositories

Top languages

Loading…

Most used topics

Loading…