Sellatrix

Sellatrix positions itself as an AI sales operating system for e-commerce and retail teams running conversations on WhatsApp, Telegram, Twilio, and a web widget. The core workflow is a fixed, ordered pipeline: every incoming message passes through topic filters, intent reads, catalog search, guided checkout steps, and an explicit confirmation gate before an order is created. The vendor states the pipeline is not an LLM free-for-all — guards short-circuit early and LLM calls execute only when the step requires one, which keeps responses fast and reduces the surface area for hallucinated pricing or missed address fields.

The differentiating architectural choice is the Client → Branch → Channel hierarchy. A single Sellatrix account can host multiple branches — each with its own configured tone, language, and LLM API key (OpenAI, Gemini, or Groq). This means a retail chain operating across three cities can run region-specific AI behavior without separate deployments or shared API rate limits. Broadcast campaigns, scheduled against imported contact lists via approved WhatsApp templates, sit inside the same dashboard as the inbound pipeline.

Sellatrix fits teams whose sales motion maps cleanly onto the browsing → variant selection → address collection → confirmation sequence. It breaks when a product catalog requires conversation logic that does not fit those steps — negotiated pricing, multi-item bundles with conditional discounts, or post-order upsell flows. Because the platform is not open-source and no self-hosted option is described on the vendor page, teams with data residency requirements or the need to modify pipeline logic at the code level will hit a hard wall. At that point, teams typically move toward a fully programmable messaging API layer combined with their own LLM orchestration.

On the integration side, the vendor describes HMAC signature verification on every incoming webhook across all four channels. LLM configuration is per-tenant and requires the customer’s own API key, so Sellatrix itself sits between your channels and your chosen model rather than providing model access directly.