Build smarter solutions with Leo AI

Applications
4
Front-ends, one backend
Design Screens
64
Production prototype
Agent Layers
12
Fully configurable
Platform Architecture

Four front-ends, one backend

An AI agent platform — companies use Leo AI to build agents that handle customer conversations across chat, voice and messaging, with human handoff when needed.

Application Used by Purpose
Leo AI Console
Leo AI platform team
Operate the whole platform: every customer org, every agent, billing, incidents, marketplace
Leo AI Studio
Customer — agent builders
Build, test, deploy and monitor their own AI agents (Northwind Retail's workspace)
Leo AI Agent Desk
Customer — support agents
Handle conversations the AI escalates to a human — with full AI context and Copilot assist
Consumer Surfaces
End consumer (shopper)
The branded chat widget embedded on the customer's site — wears Northwind Retail's brand
SOC 2
Security
GDPR
Compliance
99.9%
Uptime
24/7
Support
Applications

Four purpose-built applications

Each app targets a specific user and a specific moment in the AI-first support workflow.

APP 01 / OPERATOR
Leo AI Console
Internal panel — Leo AI's own platform team

Multi-tenant from above — sees every customer org, every agent fleet, all billing and all incidents in one place. A dense, professional internal tool optimised for fast scanning and drill-down.

Command Center — cross-customer health dashboard; platform-wide KPIs
Customers & Agents — every org, every deployed agent; drill into rows
Guardrails & Models — mandatory guardrail catalogue; LLM routing
Marketplace, Billing, Audit — blueprints, invoicing, full audit log
27 screens
Open prototype
APP 02 / CUSTOMER WORKSPACE
Leo AI Studio
AI-native agent builder — Northwind Retail

Not a plain admin tool — an AI-native product. Building an agent should feel like describing what you want and having Copilot draft it. Three agents: Retail Support EU, Order Assistant, Returns Concierge.

Leo Copilot — drafts agents in natural language; proactive insight cards
Agent Studio — 12-layer configuration workspace with NL refinement bar
Knowledge & Guardrails — RAG source management; safety policies
Quality & Insights — CSAT, conversation review, simulation testing
AI Native 21 screens
Open prototype
APP 03 / HUMAN SUPPORT
Leo AI Agent Desk
Human support workspace — Maya Chen, Northwind Retail

Not a generic help desk. Every conversation here is a continuation of an AI conversation — the human agent arrives with full AI context, what the AI tried, why it escalated.

Inbox — live handoff queue; AI summary of each escalation reason
Conversation (3-column) — context · transcript · AI handoff + Copilot
Copilot reply assist — suggests replies; agent tweak or send directly
SLA tracking, macros, team view — routing, SLA clocks, lead dashboard
Copilot 11 screens
Open prototype
APP 04 / END CONSUMER
Leo AI Consumer Surfaces
Branded chat widget — what the shopper talks to

The only Leo AI surface the end consumer ever sees — and it wears the customer's brand, not Leo AI's. Northwind Retail's teal widget. Leo AI is invisible infrastructure.

Brand-themeable — four --cs-brand* tokens drive the entire look
AI ↔ human seamless — handoff clearly signalled, conversation never restarts
Embeddable, isolated — mounts on any host page; styles scoped
Honest by design — always shows AI vs human; GDPR disclosure
Northwind theme 5 states
Open prototype
Agent Architecture

12-Layer Agent Model

Every agent on Leo AI has 12 layers. Layers 1–9 are configured by the customer in Studio. Layers 10–12 are operational systems.

01
Identity & Persona
Name, avatar, role, personality tone
02
Instructions & Behavior
Operating procedures, tone, answer length
03
Knowledge
RAG sources — websites, files, APIs
04
Actions & Tools
Function calls, integrations, tool catalogue
05
Guardrails
PII redaction, hallucination prevention, prompt injection — mandatory + optional
06
Escalation & Handoff
When and why the AI hands off to a human
07
Channels
Web chat, voice, WhatsApp, SMS, email
08
Model Layer
Task-based LLM routing — Fast / Balanced / Most-capable
09
Memory & Context
Conversation history windowing, customer profile
10
Evaluation & Testing
Simulation, eval cases, regression testing
11
Observability & Tracing
Step-by-step execution trace per conversation
12
Versioning & Deployment
Draft → staging → live, rollback, publish gates
+ Compliance & Governance — mandatory cross-cutting framework
GDPR / KVKK  ·  AI disclosure always on (non-negotiable)  ·  PII handling  ·  Audit logging  ·  Data retention windows
Technology

Tech Stack

React 18 TypeScript Vite Plain CSS + tokens.css react-router-dom lucide-react Inter + JetBrains Mono Anthropic Claude RAG + pgvector NestJS FastAPI (Python) PostgreSQL Redis WebSocket / Realtime pnpm + Turborepo Docker

v1 of each app builds against local mock fixtures — no backend required to start. Full backend stack is specified in each app's docs/ARCHITECTURE.md.