Your data is yours
Content you send to the abliteration.ai API belongs to you. We process it transiently to serve your request, then discard it. We do not keep your prompts, completions, or uploads after a request completes, and we never use them to train or improve any model.
Every request involves two kinds of data. Customer content is the prompts, messages, images, and outputs you send and receive. Operational telemetry is the counts and metadata we need to run and bill the service. Customer content is never retained; telemetry is kept only as described below.
What we store
The two kinds of data are treated very differently. Customer content never touches disk; we keep only the minimum needed to operate the service.
Never stored
- Prompts & instructions
- Input text and system prompts are processed in memory.
- Completions
- Model outputs are streamed straight back to you.
- Images & video
- Vision and video inputs are processed, then discarded.
- Conversation history
- We keep no server-side chat logs.
Retained
- Token counts
- How many tokens each request consumed.
- Timestamps & status
- Request time, model ID, and HTTP status codes.
- Billing metadata
- Invoice references and payment confirmations.
Zero retention by default
abliteration.ai is zero-data-retention by default. Some providers retain prompts and responses in abuse-monitoring logs for around 30 days unless you apply for special controls. We don't keep your content at all, so there is no payload to opt out of.
- Customer content is processed in memory and discarded when the request completes.
- Content is redacted from error and diagnostic logs before they are written.
- Operational telemetry is retained for your account's lifetime plus any legally required period, supporting billing, usage dashboards, and fraud prevention.
- You can request deletion of your account and billing records at any time.
Storage & retention by endpoint
These are the core inference endpoints. None of them store your content or use it for training, so every request is fully zero-retention.
| Endpoint | Used for training | Content stored | Zero data retention |
|---|---|---|---|
/v1/chat/completions | No | None | Yes |
/v1/messages | No | None | Yes |
/v1/responses | No | None | Yes |
Platform features
Beyond the core API, a few platform features touch data differently.
| Feature | Used for training | Content stored | Zero data retention |
|---|---|---|---|
| Web search & fetch | No | Query or URL sent to provider ¹ | No ¹ |
| Policy Gateway | No | Rules and audit metadata, no content ² | Yes |
| Training data generator | No | Datasets you generate, stored for you ³ | No ³ |
¹ Web search and fetch send your query or target URL to a third-party search provider, and their retention policies apply. See Web search & external tools.
² Policy Gateway stores your rule configuration and content-free enforcement metadata (decisions, categories, timestamps), never your prompts or completions.
³ The training-data generator on the console is covered by everything above: your inputs and the datasets it produces are never used to train our models or shared, and you can delete them at any time. The datasets are stored so you can review and export them.
Web search and external tools
Web search and web fetch are optional tools. They are off by default, enabled per request, and gated per project, so they run only when you explicitly ask for them.
When you enable them, your search query, or the URL you ask us to fetch, is sent to a third-party search provider so we can return live results from the open web. We don't store those results, but data sent to external providers is governed by their retention policies, not ours, so zero data retention does not apply to web search and fetch. If you require strict zero retention, leave web tools disabled.
EU data residency
For customers with European data-residency requirements, we process requests in the EU. With our EU data-residency option, your prompts and responses are handled on EU-based infrastructure and stay in-region. As everywhere on the platform, they aren't retained once the request completes.
Customer content stays in-region. Only non-content system data (account, billing, and usage metadata that contains none of your prompts or outputs) is handled globally to operate the service.
| Region | Processing | Content retention |
|---|---|---|
European Union Data residency | In-region | None |
United States Default | Standard | None |
Need EU data residency, a DPA, or specifics for a compliance review? Talk to us or email help@abliteration.ai.
Enterprise Key Management
Because customer content is never retained, most of your data has no at-rest footprint to encrypt in the first place. For enterprise deployments that still require customer-managed encryption, we can bring the limited operational and configuration data we do store under your own external Key Management System (KMS).
With Enterprise Key Management, that data is encrypted using keys you control and can revoke at any time. EKM is provisioned as part of an enterprise agreement and scoped to your environment. It is not a self-serve setting. Building an enterprise deployment that needs customer-managed keys? Talk to us to scope EKM for your KMS provider.
Frequently asked questions
- Do you log my prompts for debugging?
- No. Prompts and completions are processed in memory and never written to disk or to our logs. Request and response bodies are redacted before anything reaches operational logs. Only error codes, timestamps, and token counts are recorded.
- Is my data used to train models?
- Never. Your prompts, completions, and uploads are not used to train or fine-tune any model, and are not shared with third parties for that purpose.
- What do you retain, then?
- Only operational telemetry: token counts, timestamps, model IDs, HTTP status codes, and billing metadata, kept for billing reconciliation, usage dashboards, and fraud prevention. Never the content of your requests or responses.
- Does web search keep my data zero-retention?
- No. Web search and web fetch are optional and off by default. When enabled, your query or target URL is sent to a third-party search provider, whose retention policies apply. We don't store the results, but zero data retention does not apply to web search. Leave web tools disabled for strict zero retention.
- What about the training-data generator?
- The console training-data feature is covered by the same protections: your inputs and the datasets it produces are never used to train our models or shared, and you can delete them anytime. The datasets are stored so you can review and export them.
- Do you offer EU data residency?
- Yes. With our EU data-residency option, your prompts and responses are processed on EU-based infrastructure and stay in-region. They aren't retained, in line with our zero-retention default. Talk to us or email help@abliteration.ai for the details your compliance team needs.
- Can I request deletion of my data?
- Yes. Email help@abliteration.ai to request deletion of your account and associated billing records. Some records may be retained where legally required.
Questions about data handling? Talk to us or email help@abliteration.ai. See also our Privacy Policy, Terms of Service, and zero-data-retention overview.