# Jaz at AffixIO

URL: https://www.affix-io.com/jaz/
Maintainer: AffixIO / @PapaRichens

Independent explainer of JAZ research with a free, deterministic recursive-workflow browser simulation. This is not the official JAZ project or an endorsed integration.

The demo totals inputs or finds their maximum, divides work into child calls, enforces a shared call budget and a depth limit, and exports an unsigned JSON trace. It uses JavaScript, not the JAZ runtime. It makes no LLM calls, incurs no model charges and issues no AffixIO proof. Input values stay in the browser; standard website analytics remain active.

JAZ source: https://github.com/jaz-lang/jaz
Research: https://arxiv.org/abs/2609.26891

AffixIO SDK documentation: https://www.affix-io.com/sdk/
Consent scope and revocation: https://www.affix-io.com/ai-agent-consent/
Tool-call permission: https://www.affix-io.com/use-cases/api-tool-call-permission/

This page ships no native JAZ connector and reports no AffixIO benchmark result. Real JAZ execution requires a configured model and an isolated environment. Provider charges may apply outside this browser demo.


### Jaz live AffixIO API demo
Jaz also provides a separate, explicitly requested live attestation demo. The website server recalculates the sample sum or maximum, then calls https://api.affix-io.com/api/attest and /api/attest/verify. It verifies both the original payload and a deliberately altered result against the same ML-DSA-65 signature. The raw sample values go to the website server; the API receives an input digest, count and calculated result. A finite demo credential remains server-side. No customer SDK key is issued and no payment occurs. The downloadable signed receipt is separate from the unsigned browser simulation trace. Signature validity is not proof of user consent, payment authority, JAZ model execution or a zero-knowledge proof. Errors are shown as unverified, never replaced by simulated success.