Skip to content
Skeptral
Go to App

Security we can stand behind. No badges we haven't earned.

Your startup ideas and generated codebases are your most valuable assets. Here's how we handle them today, and what's on our roadmap as we grow.

View DPA Privacy Policy
The honest status of each control

No wall of badges. One track, and every control sits where it actually is today.

Read top to bottom and you read the truth: what is in place today, what we action on request, and what is still on the roadmap. The distance between them is the point. We do not paint it over.

2 in place today1 on request1 on the roadmapno badges we haven't earned
  1. 01

    Encryption in transit

    In place today

    Traffic to the platform runs over HTTPS end to end. Data lives on a single server whose volume is not separately encrypted at rest, so we don't claim that; access to production is limited to the operator and logged.

  2. 02

    GDPR and CCPA rights

    two states, told apart

    You stay in control of your data. This one isn't a single status, so we won't flatten it into one.

    ExportIn place today

    Exporting your reports is self-serve. No ticket, no wait.

    DeletionOn request

    To delete your account and associated data, email privacy@skeptral.com and we action it within 30 days.

  3. 03

    Formal certifications

    On the roadmap

    We don't yet hold SOC 2 or ISO 27001, and we won't claim badges we haven't earned. Pursuing a formal audit is on our roadmap as the team scales.

    SOC 2 ISO 27001shown as not yet earned

The badges we don't have are listed here too. Next, exactly what leaves the platform.

LLM privacy

How we work with model providers. Two things cross the line; everything else stays here.

Skeptral sends prompts to a small set of frontier model providers, currently Anthropic (Claude) and Google (Gemini), with an optional independent third judge. In production those calls travel through OpenRouter, the routing layer that relays them, so it receives every prompt in order to pass it on. We name every party that receives a prompt, and we send only what is needed to run your request.

egress.manifest// what leaves the platform, per runsigned
Sent to model providers2

the request payload, in full

  • prompt

    Your prompt for the current run

    the one thing the model is being asked to act on

  • settings

    The model and generation settings

    which model to call, and how to call it

routed via OpenRouter, arrives at
OpenRouter Anthropic Google
Never sentheld back

stays on the platform, never in a payload

  • credentials

    Your password or billing details

    credentials and payment data never reach a model

  • other_users

    Other users' data

    no one run can see another account

// nothing else is assembled into the request.
// not redacted at the edge. never sent in the first place.
Where this trust actually rests
The providers state that API inputs are not used to train their models. We rely on those terms rather than our own audit, and we will keep this page updated as our agreements and controls evolve.
We say so plainly, rather than imply an audit we do not run.Your ideas remain yours.