Trust
Trust & responsible AI
Principles
Six rules we build by
- 01
People approve what matters
AI drafts, sorts and suggests. Sending, paying, deleting or changing systems goes through an approval step or stays with a person.
- 02
Grounded or silent
Assistants answer from your data and documents. When the sources don’t support an answer, they say so instead of guessing.
- 03
Least data, keys on the server
Each AI request carries only what the task needs. API keys stay on the server and never reach a browser.
- 04
Isolation by default
Each customer’s data is scoped to its own workspace — in queries, in retrieval and in what an assistant may look up.
- 05
Limits and audit trails
Rate limits and usage caps on AI features, and logs that show what happened without storing secrets.
- 06
Your choice of model
We choose the model per task and can run on your own provider account or endpoint, so you are never locked in.
Where data goes
Which AI provider sees what
AI requests only run when a feature needs them, and carry only what the task needs.
| Where | AI provider | What is sent | Notes |
|---|---|---|---|
| LaunchOps AI | Anthropic (Claude API) | The website’s name, URL, client name and notes, and its recent check results. | Only when a user asks for an analysis. The model has no tools. Output is validated before it is stored, and logs keep request metadata only — ids, token counts, latency and status. |
| Pexpy AI | OpenAI, Anthropic or MiniMax models — or the workspace’s own provider and key | The conversation, and the records or help articles needed to answer it. | Chosen per workspace by its administrators. Each visitor’s lookups are limited to their own records.Pexpy privacy notice (opens in a new tab) |
| Client projects | Chosen with you for each project | Only the data the use case needs. | Providers, data flows and retention are documented and agreed with you before we process any of your data. |
Account, website and form data on this site are covered by our privacy notice.
In practice
What this looks like in our products
Security
- API keys and provider credentials stay on the server; browsers never receive them.
- Each customer’s data is isolated: every LaunchOps query is scoped to the signed-in workspace, and Pexpy enforces per-account row-level security in the database.
- Passwords are stored only as salted hashes. LaunchOps sign-up is by invitation, with single-use links of which only a hash is kept.
- Outbound requests — website checks, page reviews, custom AI endpoints — may only reach public addresses, and every redirect is checked again.
- Rate limits on sign-in, public forms and AI requests, plus daily or monthly AI usage caps per workspace.
- Pexpy API and MCP keys are scoped per entity and can be IP-restricted; MCP tool calls are rate-limited and logged with argument names, never values.
How the AI behaves
- Assistants answer from retrieved sources. When the sources don’t cover a question, they say so — or, for e-mail, stay silent and leave a note for the team.
- Actions an assistant proposes pass a gate: done automatically only where allowed, otherwise queued for a person’s approval or blocked.
- Our phone agent says it is an AI at the start of every call, and a person is one request away in chat and on the phone.
- LaunchOps AI is read-only: it has no tools, cannot change systems and cannot send messages.
- Structured AI output is validated before it is stored or used; anything malformed is discarded, not guessed at.
Report a security issue
Found a vulnerability in one of our products? Email hello@pxdigitalsolutions.com with “Security” in the subject. Please give us a chance to fix it before you share details.
Questions about AI and your data?
We are happy to walk your team through a data flow or a security questionnaire.
Contact us