SECURITY
Security posture
What we do today, what we explicitly do not claim, and how to report a vulnerability. Written for the security reviewer who reads this in 90 seconds.
- Report to
- [email protected]
- Disclosure window
- 90 days
Section 1: THE BOUNDARY
Security
You choose where it runs. Self-hosted, your data never leaves your environment: not in transit, not in processing, not in logs.
On the hosted marketplace, your Pack, its data and its evidence live in your isolated tenant; the only thing that leaves it is the bounded model call, to the provider you chose, and we never train on your data. We tell every buyer which mode they are getting, where the model calls go, and what leaves the tenant, before they sign.
Huitzo runs in two delivery modes: self-hosted, inside the customer's own network boundary, not a vendor cloud like AWS Bedrock; or on the hosted marketplace, in an isolated tenant per customer. Pre-built Intelligence Packs are designed to produce results in weeks, not months. Exact timing depends on the deployment and integration scope. This page is the running statement of what we do today, what we do not claim, and how to report a vulnerability.
FIG. 01. WHERE THE RUNTIME LIVES, SELF-HOSTED.
| Mode | Runtime host | Model destination | What leaves | Evidence | Operator |
|---|---|---|---|---|---|
| Self-hosted, local model | Infrastructure you control, inside your network. | A model you host inside the same boundary. | Nothing. Not data, not prompts, not logs. Packs never phone home. | The audit record is written inside your deployment and stays there. Huitzo has no copy. | Your team. Huitzo has no access to the runtime. |
| Self-hosted, external model provider | Infrastructure you control, inside your network. | The model provider you configured, such as OpenAI or Anthropic. | Only the declared prompts for the AI steps, to the provider you chose. Nothing goes to Huitzo. The record states exactly what crossed. | The audit record is written inside your deployment and stays there. | Your team operates the runtime. The provider's terms govern the prompts it receives. |
| Hosted marketplace (huitzo.ai) | Operated by Huitzo, in an isolated tenant per customer. | The model provider you chose. | The bounded model call, to the provider you chose. Huitzo never trains on your data. | Execution evidence lives in your tenant. | Huitzo operates the platform. The partner who published the Pack, or your own team, operates the Pack. |
Self-hosted, local model
- Runtime host
- Infrastructure you control, inside your network.
- Model destination
- A model you host inside the same boundary.
- What leaves
- Nothing. Not data, not prompts, not logs. Packs never phone home.
- Evidence
- The audit record is written inside your deployment and stays there. Huitzo has no copy.
- Operator
- Your team. Huitzo has no access to the runtime.
Self-hosted, external model provider
- Runtime host
- Infrastructure you control, inside your network.
- Model destination
- The model provider you configured, such as OpenAI or Anthropic.
- What leaves
- Only the declared prompts for the AI steps, to the provider you chose. Nothing goes to Huitzo. The record states exactly what crossed.
- Evidence
- The audit record is written inside your deployment and stays there.
- Operator
- Your team operates the runtime. The provider's terms govern the prompts it receives.
Hosted marketplace (huitzo.ai)
- Runtime host
- Operated by Huitzo, in an isolated tenant per customer.
- Model destination
- The model provider you chose.
- What leaves
- The bounded model call, to the provider you chose. Huitzo never trains on your data.
- Evidence
- Execution evidence lives in your tenant.
- Operator
- Huitzo operates the platform. The partner who published the Pack, or your own team, operates the Pack.
WHAT THE EVIDENCE PROVES
In the v0.1 record format, the record shows which branch of the business logic ran, which step was automated, which person approved, and the steps where no model was called at all. A signature on the record proves the bytes were not altered after the run wrote them. It does not, by itself, prove that a production event happened; it proves that the record you are holding matches what the runtime emitted.
We believe the industry needs a clear, graded standard for how AI is operated: verifiable levels of assurance an organization can climb, from basic execution logging to fully provable, boundary-enforced AI operation. Huitzo is building toward publishing that standard openly, with our own platform as its reference implementation.
Section 2: WHAT WE DO TODAY
What security controls does Huitzo have in place today?
02.1 Self-hosted
Developer tooling and the interactive quickstart (build-with-huitzo) are public at github.com/huitzo-inc. Runtime source review is available on request.
02.2 Self-hosted
The customer deploys the runtime inside infrastructure they control.
02.3 Self-hosted
The self-hosted runtime sends nothing to Huitzo, and that includes logs. Packs never phone home; it is enforced by architecture, not policy.
02.4 Self-hosted
Audit records are written inside the deployment and stay there. Huitzo has no copy.
02.5 Self-hosted
The customer chooses the model. A self-hosted model keeps every prompt inside the network; a hosted provider receives only declared prompts, and the audit record states what crossed.
02.6 Self-hosted
Postgres and Redis are reachable only on the deployment's internal network; the customer fronts the runtime with its own TLS-terminating reverse proxy.
02.7 Self-hosted
Secrets are sourced from the host environment; Huitzo does not bundle a secret store.
02.8 Hosted marketplace
Each customer's Pack runs in an isolated tenant. Huitzo operates the platform, and the tenant boundary keeps one customer's execution and data apart from another's.
02.9 Hosted marketplace
Huitzo does not train any model on customer data.
02.10 Hosted marketplace
Execution evidence, including the audit record, is written inside the customer's tenant.
Developer tooling and the interactive quickstart live at github.com/huitzo-inc.
Section 3: WHAT WE DO NOT CLAIM
Is Huitzo SOC 2 or HIPAA certified?
No. Stated plainly, here is what we do not claim today. If any item below changes, it moves to the controls list above the same week as the certification or third-party audit lands.
03.1
Not SOC 2 Type I or Type II audited.
03.2
Not ISO 27001 audited.
03.3
No published uptime SLA in either delivery mode. Self-hosted, uptime is a property of the customer's deployment; on the hosted marketplace, Huitzo operates the platform without a published SLA today.
03.4
Not independently penetration-tested.
03.5
Not HIPAA-, PCI-DSS-, FedRAMP-, or GDPR-certified in either delivery mode. Self-hosted, the runtime can be operated inside a customer's compliant environment, but Huitzo does not certify the customer's environment.
03.6
Not certified against the Huitzo standard referenced on this site. The standard is a v0.1 draft proposal, not a finished spec, and no assurance level exists as a credential today.
Section 4: DISCLOSURE
Vulnerability disclosure
- Report to
- [email protected]
- Disclosure window
- Standard 90-day disclosure window.
- Bug bounty
- Huitzo does not run a paid bug bounty program.
Section 5: QUESTIONS
Security questions
Is Huitzo SOC 2 compliant?
Does Huitzo store my data?
Does Huitzo collect telemetry, usage data, or logs from my deployment?
Which AI models can I use, and does my data go to them?
Is Huitzo HIPAA compliant?
What is your uptime SLA?
Has Huitzo been penetration-tested?
How do I report a security vulnerability?
Section 6: CLOSING
AI operating system for regulated industries
Simple by design. Infinitely scalable. Runs where your data lives.
- Product access
- huitzo.ai, the hosted marketplace, is open by invitation to teams evaluating the product.
Or email [email protected] directly.