Most AI support platforms process your customers' conversations with one vendor-chosen model, priced per resolution, on terms you can't inspect. SiteHelp is a fully managed cloud service built the opposite way: your model, your keys, your data, flat pricing — with an exit door that's always open. Everything below ships in the product today.
Each account chooses its LLM provider and supplies its own API key (stored encrypted). Use Anthropic, OpenAI, or an EU-hosted endpoint — per client, not per platform. Your conversations are never used to train anyone's models.
AI answers aren't billed per outcome, so there is no incentive to inflate "resolutions" or ration escalations. Your AI cost is transparent token spend, visible on the AI Performance dashboard.
Optional strict grounding: the assistant answers only when the knowledge base covers the question, cites the articles it used — visibly, to the visitor — and hands off to a human rather than improvise. Every reply is labeled as AI.
Card numbers and CVVs pasted into any conversation are detected and masked before anything persists — transcripts, backups and AI calls stay outside PCI scope by design.
Per-account retention windows with automatic purge, plus data-subject export and erasure tooling (GDPR Articles 15–17, DPDP-ready). Deletion means deletion, on your schedule.
Strict consent mode holds all cookies and tracking until your consent banner approves or a visitor starts a chat; AI disclosure and recording notices are built in, with consent receipts logged.
SAML 2.0 SSO with your identity provider, SCIM user provisioning, TOTP two-factor, IP allowlisting, and a full audit trail of sensitive actions.
Conversations, leads, tickets, knowledge base and audit events are yours to export at any time. No lock-in by data hostage.
Per-resolution AI pricing makes support costs unpredictable and creates a structural incentive for vendors to maximise billable "resolutions" — including tuning bots to answer confidently rather than escalate honestly. And regulators on every continent now expect chat vendors to prove consent handling, AI transparency and erasure — not just promise them. We build those behaviors into the product so every site you run on it inherits them.
We process conversation data solely on your instructions under a signable Data Processing Agreement.
A short, public list — hosting, the LLM provider you choose (with your key), mail transport, and the messaging channels you connect. See the register.
TLS in transit; credentials, API keys and provider secrets encrypted at rest.
Automated retention purge plus on-demand data-subject erasure with an audited trail.