Enterprise
Zero-retention model contracts, in plain terms
Customer code is never used to train a model, and inference runs under zero-retention contracts — commitments your legal team reads in the actual documents, not on a badge.
Zero-retention model contracts are agreements between Ciao and its model providers under which inference data is not retained by the provider, and customer code is never used to train models. Unlike consumer AI tools whose default terms may permit training on user content, Ciao makes both commitments contractual and reviewable — your legal and security teams verify the language during procurement rather than trusting a marketing page.
Published 2026-07-03 · Last updated 2026-07-03
The question every review asks first
When an enterprise evaluates any AI development tool, one question arrives before all others: where does our code go, and what happens to it there? The source code moving through an AI platform is often the company's most concentrated intellectual property — business logic, pricing rules, security assumptions. If fragments of it end up in a training corpus, there is no recall mechanism. The exposure is permanent and unquantifiable, which is exactly the kind of risk legal teams are paid to refuse.
The market has made this question harder than it should be. Retention and training policies differ between consumer and enterprise tiers of the same product, change over time, and are sometimes described in blog posts rather than contracts. A security reviewer cannot build a data-flow diagram out of reassuring sentences.
Ciao's position is short and contractual: customer code is not used to train models, and inference runs under zero-retention model contracts with the providers behind the platform. Both commitments live in documents your team can read, alongside the SOC 2 Type II report under NDA, rather than in copy that could change after you sign.
What the commitments cover
- No training on customer code — Customer code is not used to train models — not Ciao's, not the providers'. Your business logic does not become anyone's model weights.
- Zero-retention inference — Inference runs under zero-retention model contracts, so prompt and code content sent for inference is governed by contractual non-retention terms at the provider.
- Contractual, not configurational — These are terms in agreements, reviewable during procurement — not settings that a default reset or a product update could quietly flip.
- Consistent across the model ladder — Ciao runs a multi-provider model ladder with fallback, and the zero-retention posture is part of how providers participate — vendor diversity does not dilute the data commitment.
- Compatible with bringing your own model — Teams with their own provider agreements can bring their own keys and OpenAI-compatible endpoints, keeping inference under terms their counsel already negotiated.
- Attributable usage — The append-only audit trail records prompts, merges, deploys and admin actions, so what was sent, changed and shipped is reconstructable for reviewers.
How to verify this during procurement
1. Request the document set
Via the contact page, request the security pack and the data-handling terms. The zero-retention and no-training commitments are written there in contract language.
2. Put legal in front of the clauses
Your counsel reviews the actual terms — retention, training, sub-processing — during procurement, the same way they would review any processor agreement.
3. Cross-check with the audit
SOC 2 Type II reports are available under NDA, giving your team an independent account of how the platform's controls operate over time.
4. Line it up with your DPA
The GDPR Data Processing Agreement covers processing scope and sub-processors, so the model-provider layer appears in your compliance paperwork, not outside it.
5. Decide your model posture
Stay on Ciao's contracted providers, or bring your own keys and endpoints — either way, the retention posture is explicit before production work begins.
Why this matters more in an AI SDLC
A code-completion plugin sees fragments. An AI SDLC platform sees the whole delivery loop: requirements phrased in plain language, full application code, test behavior, deployment configuration. That breadth is what makes the platform useful — and it is why the data terms deserve more scrutiny here than anywhere else in your AI tooling stack. A zero-retention posture at the model layer, combined with governance at the merge layer, is what lets a security team say yes to that breadth: Guardrails applies plain-English policies and records human review, so the same loop that sees everything also evidences everything.
The posture also simplifies your own downstream story. If your organization sells software, your customers' security teams will eventually ask you the questions you are asking now; a documented no-training, zero-retention chain at the model layer gives your questionnaire answers a clean foundation instead of a stack of exceptions.
Commercially, these terms are part of the standard enterprise engagement rather than a premium add-on. Serious production programs start at USD 10,000 per year, and the data-handling terms apply to the platform as reviewed — request the security pack to start the paperwork.
What a reviewer should ask any AI development vendor
| Question | Ciao's answer | Where to verify |
|---|---|---|
| Is our code used to train models? | No — customer code is not used to train models | Contract terms during procurement |
| Is inference data retained by model providers? | Inference runs under zero-retention model contracts | Contract terms during procurement |
| Does vendor fallback change the terms? | The multi-provider ladder operates within the same posture | Security pack, on request |
| Can we use our own provider agreements? | Yes — own keys and OpenAI-compatible endpoints | Scoping with the enterprise team |
| Is there independent audit evidence? | SOC 2 Type II | Report under NDA via /contact |
Frequently asked questions
Does zero retention mean Ciao stores nothing at all?
Zero retention describes the model-provider contracts: inference content is not retained by providers, and customer code is never used to train models. The platform itself necessarily stores your project code and its append-only audit trail — that is your application and your evidence, under the terms in the DPA and security pack.
Can we get the commitments in writing before we buy?
Yes — that is the point. The no-training and zero-retention commitments are contractual language reviewed during procurement, and SOC 2 Type II reports are available under NDA. Request the security pack via the contact page to begin.
Do the terms hold when the model ladder falls back to another provider?
Yes. The multi-provider model ladder with fallback reduces dependency on any single vendor, and the zero-retention posture is part of how providers participate in it rather than a property of one favored vendor.
We already have zero-retention terms with a model vendor. Can we use those?
Yes. Bring your own keys or connect an OpenAI-compatible endpoint, and inference for your workspaces runs under the agreement your counsel already negotiated. See the own-LLM page for how that is scoped.
How does this interact with GDPR?
The GDPR Data Processing Agreement covers processing scope, sub-processors and transfer mechanisms, and the model-provider commitments sit inside that framework. Your privacy team reviews the DPA and the data-handling terms together during procurement.