The questions buyers
actually ask.

Card numbers, SSNs and API-key-style secrets are redacted on receipt, before the message goes to the AI model or the conversation record, on web chat, email, SMS, WhatsApp and the playground. Patterns covered today: US social security numbers written in the usual 3-2-4 grouping, payment card numbers, AWS access keys, Stripe live and test keys, GitHub tokens, and generic high-entropy secrets. Matched values are replaced with a labeled tag such as [CARD-REDACTED], and Satatam does not store the original except in two cases: messages are saved as received while a human operator has taken over a web chat, and for emails held by loop protection. Images are not redacted. Email addresses and phone numbers are not redacted either; they are needed to follow up.

Knowledge base files are stored per-tenant. Satatam can only read documents that belong to the tenant it is serving. Cross-tenant access is structurally blocked at the file-resolution layer.

Satatam escalates after repeated failed attempts, when a customer asks for a person, and in situations the AI judges sensitive (security, legal, threats, public reviews). The billing-dispute threshold is configurable. When it escalates, Satatam hands the thread to your team with the full transcript attached, along with the customer's last message, the resolution status and any risk flag, so nobody has to ask the customer to start over.

You configure where escalations land (Slack channel, HubSpot ticket queue, Zendesk view). Failed handoffs are flagged for follow-up.

Upload Markdown, text, HTML or PDF documents through the operator app. Each upload is versioned with a content hash, so older answers stay reproducible after you publish corrections.

Updates usually reach live Satatam within a few minutes: an ingestion worker picks up your upload on a 60-second cycle, checks the new revision against the index, and activates it. There's no retraining step. The model reads your latest version on each turn; you don't pay for retrain time, and you don't sit through a stale window.

Live today: HubSpot (CRM + helpdesk), Slack (chat alerts + threaded handoff), Calendly and Google Calendar (booking; the assistant offers open times and books the slot the customer picks), Twilio (SMS via toll-free verification, from Growth up), Stripe (subscription data and refunds), Google Sheets (analytics export), Zendesk (alternative helpdesk), and Gmail (email follow-ups sent from your own Gmail address; Gmail and Google Calendar connect with Google sign-in, everything else with an API token).

11 further providers are connectable today with an API token and marked Beta until we've verified each against a live account: Freshdesk, Intercom, Salesforce Service Cloud, Help Scout, Front, ServiceNow, Jira Service Management, Gorgias and Kustomer (helpdesks), plus Zoho and Pipedrive (CRMs).

From signed contract to production launch: typically four to seven business days. Day one to two is knowledge-base ingestion and tone calibration. Day three to four is tool wiring. Day five is shadow mode on your live conversations, where Satatam drafts replies and nothing reaches the customer until your team sends, edits or discards the draft. Days six and seven are the supervised cutover.

Go-live under your brand, with only a small “Powered by Satatam” mark left on the widget, happens when you say so.

Satatam replies in 42 languages, including Spanish, French, German, Italian, Portuguese, Dutch, Polish, Japanese, Korean, Mandarin, Arabic, Hindi, and Hebrew. It detects the customer's language from the first message and responds in kind, no manual switch required — for thirty-nine of them. Malay, Swahili and Tagalog are served but not auto-detected yet, so set the language explicitly for those.

You set the agent name, the brand color, the signature block, and the tone (Friendly, Professional, Empathetic). The end customer sees a specialist working under your brand. Everything is white-labeled except a small “Powered by Satatam” mark on the widget.

Replies typically start streaming within a few seconds. We don't publish a latency guarantee. When the upstream model is degraded, the request retries with backoff, up to three attempts in our model gateway. If none of them produce a reply, the customer gets a short message saying there was a brief connection issue rather than an open-ended spinner, the failure is logged on our side, and, when a CRM is connected and the customer has shared an email or phone number, the contact is saved there on a best-effort basis so the thread can be picked back up.