Audrex deploys grounded AI agents across your storefront — agents that resolve support, sell with context, and act within explicit authority. Every decision traced. Every action auditable.
Audrex is built as infrastructure: a multi-agent system that understands your store, acts inside defined authority, and produces evidence for every decision it makes.
Product claims come from your synced catalog. Policy answers come from your knowledge base. When retrieval finds nothing, the agent escalates instead of improvising.
Agents act through typed tools with per-agent allowlists. A sizing agent can't touch orders. Adding an agent never widens the blast radius.
Routing decision, tools called, latencies, sources cited, outcome — recorded per conversation. Debug a bad answer in minutes, prove a good one instantly, and feed what you learn back into agent instructions.
Confidence thresholds and escalation rules route edge cases to your live inbox — with the full trace attached, so your team never starts cold.
Catalog sync, theme-extension widget, app-proxy security, order workflows — native from day one, not bolted on.
Automation rate, handling cost saved, and recommendation-to-order attribution — your CX becomes a measurable line item, not a cost center.
Product webhooks keep embeddings fresh. Change a price, publish a product, update a policy — agents know within moments.
A triage agent reads every inbound message and routes it to the right specialist — each with its own persona, instructions, and a scoped set of tools it is allowed to use.
Audrex ships with production-ready workflows for the customer moments that drive retention and revenue — plus a builder for whatever your team ships next.
Looks up real order data by number and email — fulfillment status, tracking number, and carrier link, not a guess.
Delivery windows quoted from your real shipping policy, by destination — no generic "3-5 days" for every country.
When a shopper shows high intent but doesn't buy, an automation can tag them, alert your team on Slack, or issue a discount code — your rules, your trigger.
Answers pull from your own sizing guide and policy documents — the same knowledge base your support team would use.
Recommendations are checked against live inventory from your catalog — nothing suggested that's out of stock.
Records the return request against the real order and routes it to your team for approval — no fabricated policy exceptions.
When retrieval comes up empty or confidence drops, the agent hands off to your live inbox with the full trace attached.
Audrex agents share a single memory and policy layer — no matter where the conversation happens. Add channels without rebuilding your knowledge or your persona.
Enterprise CX isn't about letting a model improvise. It's about defined authority, verifiable grounding, and a record of every decision. That's the contract Audrex is built on.
Retrieval runs before generation. Traces show the exact documents and products behind every answer, so hallucinations are detectable — and by design, rare.
Every capability is a typed tool granted per agent. Nothing is implicit, nothing is inherited. Review the full authority map at any time.
Routing, tool calls, latencies, sources, outcome — one immutable record per conversation. Compliance reviews stop being archaeology.
Confidence thresholds, escalation rules, a live inbox, and one-click takeover. The agent works for your team — never around it.
Legacy chatbots deflect. Human teams don't scale. Audrex is the third path: agents that resolve, act, and prove it.
Audrex is built for the moment a Shopify team outgrows FAQ macros and a shared inbox.
A support queue that only grows — every question answered from memory, nothing grounded in the actual catalog or policy, and no record of why an answer was given.
A catalog too large for a single person to know cold — shoppers ask specific questions about stock, variants, and sizing that a generic chatbot script can't answer.
A compliance or ops lead who needs to know exactly why an AI system said what it said — not a black box, a record that can be reviewed and defended.
A brand voice that can't sound like a generic bot — every automated response needs to read like it came from the team, not a template.
A single inbox that's stopped scaling — escalations arrive with no context, and every handoff means starting the investigation over from scratch.
A founder still answering shopper messages at midnight — coverage shouldn't depend on someone being awake.
← swipe for more →
Start small, scale with resolutions. Every plan includes the full platform — architecture editor, traces, and human handoff.
One fully grounded agent that handles support and sales end to end.
For stores whose support volume or catalog complexity outgrows a single agent.
Both plans start with a 14-day free trial through Shopify Billing — no separate free tier, no card charged until the trial ends.
Install from Shopify, sync your catalog, load your policies. Audrex maps a recommended agent architecture for your store automatically.
Tune personas and tool scopes in the architecture editor. Pressure-test agents in the playground and review every trace before launch.
Enable the storefront widget. Watch automation rate, resolution time, and revenue attribution compound as the system learns your store.
Support stops being a cost center the moment you can see every decision the agent made. That's the shift Audrex is built around — a system you operate, not a queue you survive.
A chatbot generates text. Audrex operates a system: a triage agent routes each message to a specialist with scoped tools, answers are grounded in your catalog and knowledge base, actions are typed and auditable, and every decision is recorded in a replayable trace. It's the difference between a script and an operating model.
Retrieval runs before generation — product and policy answers are drawn from your synced data. When retrieval returns nothing relevant, the agent declines to improvise and escalates to your team. Traces show retrieved sources for every answer, so hallucinations are detectable and, by design, rare.
Complete control. You define each agent's persona, instructions, and exact tool allowlist. Confidence thresholds and escalation rules decide when humans take over. The live inbox gives your team full visibility, and any conversation can be taken over in one click.
Install from the Shopify admin, sync your catalog (automatic), add policy documents to the knowledge base, and test in the playground. A first production agent typically ships within days. The architecture editor lets you evolve the system afterwards without redeploying anything.
Storefront chat via theme extension and the embedded Shopify admin experience are available today. Email and social channels, plus expanded human-takeover workflows, are the next layer of the roadmap.
Each store's data is isolated to its own tenant and used to serve that store. Traffic is encrypted in transit on every hop, storefront requests are verified through Shopify's signed app proxy, and internal APIs require authenticated service keys. Export and deletion are available on request.
Pricing, security review, migrating from another tool — a real person answers, not a ticket queue.
Connect your store and see what Audrex resolves in the first five minutes — with every decision on the record.
Connect Shopify →