The big picture
Parlel sits as an invisible middleware layer between your customers and your backend systems. Customers interact through AI tools they already use — ChatGPT or Claude — or through Parlel-powered interfaces embedded directly in your own product. Parlel handles everything in between: what the AI knows, what it can do, who it’s talking to, and what rules apply. You configure it once. It works across all interfaces automatically. Your existing systems stay unchanged.How customers interact with Parlel
Parlel supports two categories of interface: AI tool integration — customers use Parlel through ChatGPT or Claude, the AI tools they already have. Parlel connects to these platforms via API, making your configured knowledge, actions, and rules available inside the customer’s existing AI workflow. No new tool to learn, no new login. Parlel-native interfaces — for companies that want a branded, controlled experience, Parlel offers two native surfaces. The parlel.ai chat is a standalone interface at parlel.ai. The company web widget is an embeddable overlay you deploy on your own site or app with a single script tag — it opens as a full-viewport chat branded as Parlel, connected to your company. The widget is fully isolated via Shadow DOM so it never conflicts with your existing styles.The three capability layers
Knowledge layer
Ensures the AI always answers from your content — your products, policies, pricing, and FAQs — never from generic training data. When a customer asks a question, Parlel retrieves the most relevant content from your knowledge base and provides it to the AI as the authoritative source. If the AI isn’t confident in an answer, it escalates cleanly rather than guessing.Commercial layer
Gives you control over how conversations flow and what commercial rules apply. Customer journeys let you define structured paths — an upgrade flow, a cancellation retention flow, an onboarding sequence. Campaigns and mandates let you surface offers and negotiate within boundaries you set. A transaction basket tracks what the customer has agreed to during the conversation, ready for action execution. Everything fires at the right moment in the right conversation context.Authentication layer
Verifies customer identity progressively through four tiers — from anonymous to OTP-verified. A customer starts with no identification and gains access to more sensitive actions as they verify. Friction scales precisely to what’s being asked: general questions require nothing, account changes require verified identity. See Authentication for the full tier model.The execution layer
Action layer
Translates customer intent into real transactions against your backend systems. Actions are configured flows — each one defines what data to collect, what validation to apply, what API to call, and how to handle errors. Every action execution is logged with a full audit trail.Integration layer
Connects Parlel to your existing systems via pre-built connectors and generic REST integration. HubSpot, Salesforce, Stripe, Zendesk, and custom APIs are all supported. Your systems don’t change — Parlel connects to them as they are.The configuration portal
Everything above is configured through the Parlel portal at parlel.ai/portal. Knowledge, actions, journeys, authentication rules, commercial rules, and integrations are all managed in one place. Configure once, deploy across all interfaces automatically.What happens when a customer sends a message
Every message goes through the same sequence:1
Session resolved
Parlel identifies the session and determines the customer’s current authentication tier — anonymous, identified, or verified.
2
Context assembled
Three things happen in parallel: relevant knowledge is retrieved from your knowledge base, customer data is loaded from your CRM, and applicable commercial rules and journey state are evaluated.
3
AI responds
All context is provided to the AI with your configured rules. The AI generates a response grounded in your knowledge, appropriate to the customer’s identity, and compliant with your commercial rules.
4
Action executed
If the customer confirmed an action, it is executed against your backend systems, validated, logged, and a transaction record is created.
5
Response delivered
The customer receives the response inside the interface they’re using — their AI tool or your embedded widget. Nothing changes on their end.
Where to go next
- Understand the Knowledge Base — how your content becomes AI knowledge
- Understand Actions — how customers take real actions through AI
- Understand Journeys — how you control conversation flow
- Understand Authentication — the tier model and verification flow
- Start configuring with Getting Started