Dialog

The agent acts in systems the vendor does not own, and there is a route to add your own.

Dialog installs as a Shopify app and embeds through an SDK and CDN script that expose cart management and product fetching, so the agent adds to cart in the merchant's store rather than only advising, and Smart Reengagement captures a shopper's contact details in conversation and pushes them into the brand's connected email or CRM system.

Developer mode adds external API calls, webhook triggers and custom rules, so the set of connections is not fixed. Reach extends onto external assistants including ChatGPT, Perplexity and Gemini, where the brand's own agent answers.

Sourceapps.shopify.com/dialog-ai-sales-assistant, npm SDK and askdialog.com/agentic-commerceread 2026-09-01

Runs guided conversational sales flows that qualify a shopper's needs, recommend products with encoded recommendation logic and sequencing, and drive toward purchase, but not complex multi step or multi agent orchestration.

Sourceaskdialog.com/agentic-commerceread 2026-10-01

A maintained structure over the brand's own material, described as encoding rather than prompting. Dialog ingests the live catalog alongside the brand's documentation and content, and the platform page frames the third encoding dimension as expertise: product knowledge, sales logic and recommendations, built on how the brand's best sellers advise.

That store persists, follows the catalog as it changes, and is read at conversation time to answer product questions and drive recommendations, with dynamic FAQs assembled from real shopper questions on top of it. The vendor says the agent is trained on the brand's catalog and tone, though what is documented is ingestion and retrieval rather than fine tuning.

Sourceaskdialog.com/agentic-commerceread 2026-09-01

A configured control surface with escalation, published as one of the platform's three pillars. Dialog encodes a brand across identity, responsibility and expertise, and names the responsibility dimension explicitly as guardrails, policies and escalation, describing them as the rules you set for your customers.

The agent hands off under conditions the brand defines, and the same interface lets a team tune how it answers, recommends and escalates. The guardrails are named as a pillar rather than documented in detail, and no per action gate, permission model or escalation configuration is published.

Sourceaskdialog.com/agentic-commerce, brand encoding sectionread 2026-09-01

The homepage shows an AICPA SOC 2 badge alongside a GDPR badge. Shopify's data governance belongs to Shopify rather than Dialog. No named customer facing control (single sign on, role based access control, audit logging or retention configuration) is published, and no SOC 2 report or trust portal is publicly available.

Sourceaskdialog.com homepage badge rowread 2026-09-15

The transcript is the record, and there is nothing beyond it. Dialog says there is no black box and lets a team review past conversations, and Conversation Insights logs and analyses every question a shopper asks to surface recurring objections and gaps, with the SDK emitting positive and negative feedback events per response.

For a conversational agent the transcript is a genuine record of what it did. No audit log, run history, decision trace or record of which catalog item or rule produced a given answer is documented, and the analytics layer describes shopper behavior rather than agent execution.

Sourceaskdialog.com/solutions/conversation-insights and the npm SDK feedback eventsread 2026-09-01

Per-shopper state outlives the conversation: Smart Reengagement identifies hesitant shoppers on what they asked but did not buy and carries that into a follow-up, so the exchange is retained and read back rather than discarded at session end. No memory feature is named, scoped or given a lifetime, and the persistent brand encoding of identity, guardrails and expertise is configuration rather than memory.

Sourceaskdialog.com/solutions/smart-reengagement and askdialog.com/agentic-commerceread 2026-09-01

Cloud only, with nothing the customer selects. Dialog is delivered as a hosted service embedded through a script, an npm package or a Shopify app, and no region choice, data center disclosure, VPC, private tenancy or self hosted option is published. The developer mode option for an agent by country is audience segmentation, not data residency.

Sourceaskdialog.com/agentic-commerce and npmjs.com/package/@askdialog/dialog-sdkread 2026-09-01

A named set of prebuilt agent modules, each with its own product page in the vendor's navigation: AI Personal Shopper for conversational discovery across the catalog, AI PDP Assistant for answering questions and handling objections on the product page, Smart Reengagement for identifying hesitant shoppers and capturing contact details, and Conversation Insights for turning conversations into merchandising and content signals. A brand switches these on rather than assembling them.

Sourceaskdialog.com solutions pages and site navigationread 2026-09-01

The merchant's own customers start the agent's work rather than the merchant's staff. A shopper's question on the homepage, a product page or search, or inside ChatGPT, Gemini, Claude or Perplexity, wakes the agent with no one at the merchant starting it, and a Smart reengagement module reaches back out to shoppers.

Sourceaskdialog.comread 2026-10-01

No model provider is named on the vendor's current pages, and no customer selectable model or multi provider routing is documented; ChatGPT, Gemini and other assistants appear as distribution surfaces rather than model choices.

Sourceaskdialog.com/agentic-commerceread 2026-09-01

A published SDK plus a documented developer surface.

The TypeScript SDK ships on npm as @askdialog/dialog-sdk, instantiated with an API key, with a CDN build exposing DialogSDK on the window for teams not using a package manager, and covering assistant interactions, product fetching, cart management, appearance customization and feedback events such as userSendPositiveFeedback and userSendNegativeFeedback.

Alongside it the platform page documents a developer mode that triggers external API calls, injects custom rules, fires webhook triggers and manages multiple agents per country, catalog or persona.

Sourcenpmjs.com/package/@askdialog/dialog-sdk and askdialog.com/agentic-commerceread 2026-09-01

A documented optimization loop the customer runs, about the agent. The platform page heads a section Continuously test, evaluate and improve, and describes the loop in four steps: review past conversations, understand how users behave, fine-tune how the agent answers, recommends and escalates, and improve performance, all from one interface, with the vendor framing it as no black box. What is being tuned and measured is the agent's own behavior rather than a downstream business metric. A/B testing of agent versions is not documented.

Sourceaskdialog.com/agentic-commerceread 2026-09-01

Dialog is an embedded conversational agent (including photo based multimodal input) but performs no browser automation or general computer use.

Sourceaskdialog.com/agentic-commerceread 2026-10-01