# AICLUDE (long-form)

> AICLUDE (aiclude.com) is the enterprise AI agent platform. Design, deploy, and operate AI agents with multi-stage pipelines, Graph N-hop Agentic RAG, real-time voice and lipsync avatars, kiosks and signage. Same Docker images run as Cloud SaaS, on-premise, or air-gapped. AICLUDE is unrelated to Apple iCloud.

This long-form llms-full.txt collects the complete content surface of aiclude.com: every product, every industry solution, every documentation entry, and every published blog post, all keyed to canonical URLs. Use it as a single ingestible artifact for AI search and citation systems.

## Section 0. What is AICLUDE?

AICLUDE is an Enterprise Agent OS. Pronounced "AI-clude" (eye-clood), the name is a portmanteau of "AI" and "include": the platform that includes AI in every business process. AICLUDE Inc. (아이클루드 주식회사 / アイクルード) is the company; aiclude.com is the home of the platform. Founded 2024. Areas served: KR, JP, US, Worldwide.

AICLUDE is a software platform brand. It is not Apple iCloud, has no relationship with Apple Inc., and does not offer consumer cloud storage. Apple iCloud (i-Cloud) is Apple Inc.'s personal cloud storage and sync service; AICLUDE is an enterprise AI agent platform sold to organizations.

Three products share one CLU Agent OS core:

- CLU Agent Platform (CAP) for agent design and orchestration.
- CLU Agent Station (CAS) for multi-device frontends including kiosks and signage.
- CLUE for an enterprise-grade AI database (RAG plus vector, knowledge graph, and keyword search in one engine).

A fourth product, ASVS (AI Agent Security Vulnerability Scanner) at vs.aiclude.com, pre-scans Skills and MCP servers before they reach your agents.

Slogan: "Include AI in everything you do."

## Section 1. Products

### 1.1 CLU Agent Platform (CAP)

Canonical: https://aiclude.com/products/cap

The flagship multi-agent platform. CAP is the enterprise agent platform for knowledge workers: visual pipelines, a Knowledge Hub, 166+ tools, and a Multi-LLM router on one screen. The same agent owns chat, email, web, and social end-to-end.

Core capabilities exposed on the product page:
- Visual pipeline designer. Compose multi-step pipelines no-code, with a different LLM, tool, and verifier for each stage.
- Knowledge Hub with 6 source types. PDFs, URLs, FAQs, email, social, and SaaS, embedded once so RAG grounds every answer.
- 166+ tools, MCP-native. Slack, Gmail, Calendar, Browser, Sandbox and more, all callable as functions, plus full MCP support.
- Multi-LLM router. OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, xAI, self-hosted vLLM, and any OpenAI-compatible endpoint.
- Omnichannel deployment. Web widget, Slack, Discord, REST API, MCP server, signage, and 10+ channel surfaces.
- Enterprise security. Physical tenant isolation, AES-256-GCM PII encryption, per-group dedicated encryption keys.
- Same binary on-prem. The same Docker images you run on Cloud also run on customer infrastructure.

Live demo (five-stage agent turn): walk from input through RAG, generation, verification, and delivery, and KPI cards plus four SNS copies arrive together.

Personas: knowledge workers, product and IT teams, marketing, and operations.

Repeated workflows on CAP: internal IT helpdesk, marketing operations, customer support assistant, sales research and proposals.

### 1.2 CLU Agent Station (CAS)

Canonical: https://aiclude.com/products/cas

CAS is the multi-device frontend for AICLUDE agents. It runs as a browser app on kiosks, signage, mobile (9:16), tablets (16:9), PCs, and the web. CAS detects passersby with a camera, greets them in their language, converses in real-time voice with lipsync avatars, and hands off to a human agent through the Takeover flow.

Capabilities:
- AI Concierge with STTS (speech-to-speech) voice.
- Smart digital signage with auto video.
- Camera pedestrian detection and engagement.
- Avatar lipsync.
- Attention and gaze analytics.
- Weekly PDF report.
- Mobile video-call mode: the persona's face renders with lipsync over real-time audio for a phone-like interaction.

Hardware bundles (display, camera, microphone) are quoted separately. Optional.

Personas: retail, hospitality, public and exhibition spaces, healthcare intake.

Pricing concept: a "station" is the billing unit. One signage or one AI concierge running at a store equals 1 station. A store running both signage and concierge counts as 2 stations. A "device" is the screen that actually displays the station, unlimited. Mirroring one station to many devices (entrance signage plus mobile sync, for example) is at no extra cost.

### 1.3 CLUE

Canonical: https://aiclude.com/products/clue

Enterprise AI Database. Tagline: "Every answer leaves a clue. Stack docs, code, images, and audio on a single database, query four ways with the same API."

Capabilities:
- 4 search modes in one engine: vector, keyword, knowledge graph, multi-hop graph traversal.
- 30+ file formats auto-ingested.
- Vector + graph DB on a single PostgreSQL-native core (pgvector, AGE Graph, VectorChord-BM25, and RAG variants).
- Per-tenant data isolation.
- Pinecone-compatible SDK so existing code drops in.
- Sharded cluster, failover HA.
- Standard topology: 1 control + N data PG with the "clu" schema accepted (Supabase-like).

Personas: AI platform teams, enterprise IT, data engineers.

### 1.4 ASVS (AI Agent Security Vulnerability Scanner)

Canonical: https://vs.aiclude.com/

ASVS continuously pre-scans the global Skill and MCP server ecosystem. Thousands of packages from npm, GitHub, Smithery, and more pass through:

- 7 parallel scan engines: SAST, SCA, Tool Analyzer, DAST, Permission, Behavior, Malware.
- 10 AI-specific threat categories: Prompt Injection, Tool Poisoning, Command Injection, Credential Harvesting, Data Exfiltration, Supply Chain, Permission Escalation, Malicious Behavior, Obfuscated Code, Backdoors.
- 0-100 Risk Score across 5 severity levels.
- SBOM auto-generated.
- Scan Search, Risk Reports, Connected Agents, MCP / Skill integration.

Deployment: Docker + gVisor isolation, vendor-neutral, Oregon-only operation, portable consultant box.

## Section 2. Core Platform Capabilities

- 8-stage agent pipeline: Input enrichment → Intent understanding → Pre-validation → Execution planning → Fast response → Post-validation → Auto-correction → Telemetry.
- 4 RAG variants: Classic RAG, Graph RAG, Agentic RAG, Hybrid RAG.
- Graph N-hop Agentic RAG: papers, patents, SOPs, and in-house experiment data split by proposition, retrieved via vector + keyword hybrid, then walked across graph edges for 2-3 hop reasoning.
- Hybrid execution engine: deterministic skills for repeated work, autonomous LLM execution for novel requests.
- Self-evolving Skills: skills measure their own outcomes and iterate on weak spots.
- 166+ built-in tools: documents, image, video, browser, scheduler, code sandbox, DB query, data analysis.
- Function Calling Tool Registry: register customer APIs as external tools, invoked by natural-language requests.
- Multi-LLM router with Circuit Breaker: OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, xAI, self-hosted vLLM, any OpenAI-compatible endpoint. Different LLM per pipeline stage (input, understanding, generation).
- BYOK across all major vendors, or AICLUDE-managed keys with the AICLUDE Token Fee (ATF) tier.
- Real-time voice + lipsync avatar with STTS, native multilingual Korean / English / Japanese.
- 4 image providers + 3 video providers + voice synthesis + lipsync + background audio + AI landing builder, orchestrated on a DAG autonomous execution engine.
- Persona RAG + Brand RAG: agents stay on persona and brand voice across channels.
- Workspace integrations: Email, Calendar, Drive, Slack, Discord, SNS, 9+ channels via one agent core.
- Fortress Pipeline: the integrated security envelope around the 8-stage core; every response carries source citations and validation traces.
- 9 AI security guards: prompt injection, tool misuse, data leakage, RAG poisoning, chain attacks, memory poisoning, DoS, output tampering, supply chain, mapped to OWASP LLM Top 10.
- AES-256-GCM PII encryption at rest, search-while-encrypted indexing across 13+ PII tables and 40+ fields.
- Physical multi-tenancy: each tenant gets its own table set; per-tenant KeyVault separates encryption keys.
- Audit Log: every agent action and data access recorded for compliance, appeals, and incident review.
- Same Docker image for SaaS and on-prem; encrypted container images via a private registry; full air-gap support.

Platform numbers as published:
- Core Pipeline Stages: 8+.
- Built-in Tool Library: 100+ (the marketing copy quotes 100+; the comparison table lists 166+).
- Enterprise Tenant Isolation: 100%.
- Omnichannel Deployments: 10+.

## Section 3. Solutions by Industry

### 3.1 Manufacturing

Canonical: https://aiclude.com/solutions/manufacturing

Headline: from process data to R&D documents, an AI agent layer atop an encrypted factory.

Premise: factories already run MES, DCS, PLC, and digital twins. AICLUDE does not replace that OT stack; it rides on top of it as an agent layer covering unstructured data, regulatory docs, research knowledge, and a worker copilot. Encrypted on-prem container image deployment, physical multi-tenancy, and 9 AI guards are built in from day one, so sensitive process data never leaves the internal network.

Cases:
- Predictive maintenance + on-floor copilot. Vibration, temperature, action history, and SOPs unified through Graph N-hop Agentic RAG. On anomalies the hybrid engine retrieves similar cases and auto-generates an action guide; operators query in natural language from mobile or tablet.
- Process-quality image analysis reports. Vision detection (defects, cracks, contaminants) wired to LLM explanation and in-house report templates trained via RAG; Fortress Pipeline 8 stages filter hallucinations; Audit Log captures the entire inspection and decision flow.
- R&D document and patent Graph N-hop Agentic RAG. Papers, patents, and in-house experiment DBs split via Graph N-hop, hybrid vector + keyword search, natural-language query to comparison table and report draft; native KO / EN / JA enables simultaneous global search.
- Global patents and technical docs, multilingual. Regulatory source texts from 4 jurisdictions (FDA, EU MDR, NMPA, MFDS) learned together; competitor patent DB connected for auto-monitoring and summary reports; 4 image providers integrated to automate brochures and technical drawings.

Tile highlights: Graph N-hop Agentic RAG, encrypted container images, Vision analysis, 166+ tools.

### 3.2 SI / System Integration

Canonical: https://aiclude.com/solutions/si-integration

Headline: different security demands per customer, one platform, one Docker image.

Premise: SI engagements must satisfy different security, regulatory, and data policies for every customer. AICLUDE flips between Cloud and Self-hosted deploy modes from one Docker image and isolates customers with physical multi-tenancy. 9 AI guards, ASVS, and Fortress Pipeline are built in from day one, so SI partners never have to redesign the security layer per customer.

Cases:
- Accelerated customer AI agent rollout. No-code agent builder, 166+ Function Calling tools pre-loaded, external API tool registration to plug customer APIs in like extensions, persona / pipeline / model tuning live from the admin UI.
- Same image for on-prem and SaaS. Toggle Cloud and Self-hosted from the same Docker image; deliver encrypted images via private container registry; customer install scripts automate repeat deployments; one release updates both simultaneously.
- Multi-tenant customer management. Physical multi-tenancy, hundreds of tenant tables architected, AES-256-GCM with search-while-encrypted index, per-customer KeyVault, Audit Log per customer.
- AI governance standard platform delivery. 9 AI guards + 8-stage Fortress Pipeline shipped as a standard, ASVS for external skill verification, 5-vendor LLM router with auto failover, security-designed AI standard delivered to customers.

Tile highlights: deploy mode switch, physical multi-tenancy, encrypted container images, no-code builder.

### 3.3 IT Services

Canonical: https://aiclude.com/solutions/it-services

Headline: fragmented knowledge, endless on-call, an AI agent that follows your team.

Premise: IT and software orgs scatter knowledge across wikis, Slack, Drive, and DBs while on-call, incident response, and onboarding rely on tribal knowledge. AICLUDE links heterogeneous sources via Graph N-hop Agentic RAG and learns repeating org work patterns through a hybrid engine.

Cases:
- Unified internal knowledge search AI. 3 data connectors (S3, Google Drive, DB) auto-link sources; Slack bot and web widget for instant natural-language Q&A; Audit Log captures who accessed what.
- Issue triage and prioritisation AI. Org issue history learned via Graph N-hop Agentic RAG; LLM auto-classifies type, suggests owner, decides priority; Function Calling updates the ticket DB.
- Developer onboarding copilot. Internal architecture docs, coding conventions, and API specs learned; new devs ask in natural language; hybrid engine auto-promotes repeating question patterns into skills; reach from IDE and terminal via Slack bot, web widget, MCP.

Tile highlights: Graph N-hop Agentic RAG, hybrid execution, Slack and MCP omnichannel.

### 3.4 Entertainment & Media

Canonical: https://aiclude.com/solutions/entertainment-media

Headline: from a product line to an AI-hosted broadcast, image × video × lipsync in one pipeline.

Premise: content, ads, and broadcasting are stitched together by people. A single image or video generator cannot connect brand consistency, multilingual output, multi-format delivery, and performance feedback. AICLUDE wires 4 image providers, 3 video providers, voice-synthesis (TTS) AI, lipsync AI, and the AI landing builder into a single DAG autonomous execution engine.

Capabilities:
- 4 image providers. Multiple image AI models in parallel; prompt + negative + aspect ratio auto-conversion.
- 3 video providers. Multiple video AI models in parallel; auto-generate script → scene → video.
- AI video host (TTS + lipsync). Voice synthesis + lipsync + background audio + video edit AI compose AI shopping and broadcast hosts.
- 5-stage AI landing builder. Planning, design, code, images, editing, with drag-and-drop visual editor.

Tile highlights: 3 video providers, lipsync AI, AI landing builder.

### 3.5 Retail & Distribution

Canonical: https://aiclude.com/solutions/retail-distribution

Headline: web, app, and store as one agent. From discovery to checkout without seams.

Premise: retail and commerce face online / offline disconnects, no conversational shopping, no in-store AI guidance, and manual campaign loops. Existing AI builders stay on web alone with no kiosk, signage, or STTS voice support; simple FAQ chatbots cannot actually execute inventory, order, or payment APIs.

Capabilities:
- Function Calling Tool Registry. Register inventory, coupon, order, payment, and shipping APIs as agent tools; external APIs run from a natural-language request.
- Agent Takeover + Coaching Mode. Human intervention via agent takeover; real-time coaching; human agents step in naturally on top of AI responses.
- Native Kiosk & Signage. Unified kiosk and signage operations built into the product.
- STTS Voice Conversation. Real-time speech recognition, AI reply, and voice response in one cycle. Native multilingual (KO / EN / JA).
- Persona-Based Hyper-Personalization. Recommendations based on customer preferences and history via the persona and customer context engine.
- DAG-Based Autonomous Campaign Loop. DAG autonomous execution engine runs segment → content → delivery → performance → retargeting end-to-end.

Live deployment: STNET × AICLUDE "Catch Sales" Kiosk PoC in Kagawa, Japan, ran Dec 2025 - Mar 2026. Two kiosks detected more than 6.4 million pedestrians; nearly two of every three responded when the AI spoke first.

### 3.6 Education

Canonical: https://aiclude.com/solutions/education

Headline: every learner at their own pace, an AI tutor personalizes learning while instructors automate assessment.

Premise: education simultaneously faces individual pacing differences, instructor lecture / assessment workloads, narrative grading time, and demand for multilingual learning content.

Capabilities:
- Graph N-hop Agentic RAG over lectures, textbooks, past exams, rubrics. Cites evidence for every learner question.
- Fortress Pipeline 8 stages. Captures learner intent, verifies, generates answers, and cites sources.
- Instructor Review Takeover. Applies the agent takeover flow as an instructor review flow: AI grades, instructor finalizes.
- Audit Log. Records all learner and instructor activity for grade appeals and audits.
- Native KO / EN / JA. Multilingual learning content and assessments simultaneously supported.
- PII Protection. Learner PII encrypted with AES-256-GCM plus physical silo isolation.

Live deployment: the AI Tonbo Teacher case at Tonbo Kingdom, Kochi, Japan, March 2025. Sixty-plus years of the real Tonbo teacher's printed books and dragonfly photographs were scanned into the knowledge base, and his actual face and voice were cloned into a digital-human AI avatar, a Classic-RAG-era instructor AI that answers with both text and scanned source images.

### 3.7 Finance & Security

Canonical: https://aiclude.com/solutions/finance-security

Headline: in finance, security is the baseline. Fortress Pipeline, 9 AI guards, and AES-256-GCM are built in.

Premise: finance simultaneously demands vast regulatory texts, PII protection obligations, and strict AI governance and audit. AICLUDE bakes in 8-stage Fortress Pipeline, 9 AI security guards, AES-256-GCM with search-while-encrypted indexing across 13+ PII tables and 40+ fields, physical multi-tenancy, ASVS skill validation, Audit Log, and a 5-LLM router with automatic failover.

Capabilities:
- Fortress Pipeline 8 stages. Physical multi-tenancy → AES-256-GCM encryption → multi-factor auth (OAuth + OTP) → Audit Log → on-prem deployment, all wrapped together.
- 9 AI Security Guards mapped to OWASP LLM Top 10.
- AES-256-GCM + search-while-encrypted index. Applied to 13+ PII tables and 40+ fields.
- Physical Multi-tenancy. Per-tenant physical table separation. Full isolation by department or affiliate.
- Audit Log. Full logging of resource type and action history; instantly answer "who, when, what data" for any audit.
- ASVS Skill Validation. External skills and tools must pass the ASVS security scanner (Safe / Warning / Blocked) before registration.

Tile highlights: Fortress Pipeline, 9 AI Security Guards, AES-256-GCM, ASVS.

### 3.8 Telecom

Canonical: https://aiclude.com/solutions/telecom

Headline: branches, call centers, and B2B sales as one AI agent. From plan recommendations to contract, without seams.

Premise: telecom has hundreds of plans, device lineups, T&Cs, and promotions updated weekly, while branches, call centers, and B2B sales each run separate systems. Existing AI chatbots stop at simple FAQs and cannot actually execute activation, number portability, or plan-change APIs.

Capabilities:
- Real-time search across plans, devices, T&Cs. Hundreds of plans / devices / options / T&Cs / promotions retrieved instantly via Graph N-hop Agentic RAG; auto-retrained as weekly changes land.
- Call-Center Agent Takeover and real-time coaching. Hand off seamlessly from agent to human; real-time coaching and response approval suggest scripts and discount options to the agent.
- Offline Branch Kiosk & Concierge. Native kiosk and signage + STTS voice; waiting customers self-serve plan comparison and device consultations; context is handed off to the human agent.
- B2B Sales RFP & Proposal Automation. Analyze enterprise requirements → draft tailored pricing and circuit plans → build competitor comparison; sales reps only review and revise.
- Function Calling Real Service Changes. Activation, number portability (MNP), plan changes, and add-on enrollment registered as external API tools; the agent executes the real API after customer consent.
- Customer Persona-Based Recommendations. Persona and customer context engine auto-recommends the optimal plan based on usage patterns, family composition, and history.

Tile highlights: branch kiosk, call-center takeover, B2B RFP automation.

## Section 4. Deployment Options

### 4.1 Cloud SaaS
- Zero ops; physical tenant VMs from Pro and up. Starter uses a shared tenant pool with per-tenant resource quotas.
- Same Docker images as Self-hosted.
- Multi-region failover included by default in Scale and Enterprise.
- Audit logs stored automatically in object storage. Pro: 90-day retention. Scale and Enterprise: unlimited.

### 4.2 Self-hosted (On-premise)
- Encrypted container images delivered through a private registry.
- License server check-in with 7-day offline grace.
- Same release as Cloud.
- Customer-side KeyVault for encryption-key separation.

### 4.3 Air-gapped Enterprise
- Full offline deployment available for Self-hosted Enterprise.
- Model ONNX packages plus encrypted image bundle delivered physically.
- No outbound connectivity required.

## Section 5. Security & Compliance

Security foundations:
- Physical tenant isolation. Dedicated VMs and separated database tables per tenant. Not row-level: infrastructure-boundary isolation.
- AES-256-GCM PII encryption. Sensitive fields encrypted at the application layer with authenticated GCM mode. Keys rotated via KeyVault.
- Search-while-encrypted index. Search encrypted fields by exact match without ever decrypting. Industry-standard secure indexing.
- Per-group dedicated encryption keys. Independent cryptographic keys per group; compromise of one group cannot expose others. Encrypted key backup.
- Encrypted container images. Docker images encrypted layer-by-layer with per-customer keys. Supply chain signed, tamper-evident.
- Multi-layer AI security guards. Prompt injection, knowledge poisoning, and PII leakage defended at platform level; aligned with OWASP Top 10 for LLM.

ASVS Skill & MCP Safety Scanner:
- Continuously pre-scans the global Skill and MCP ecosystem: thousands of packages from npm, GitHub, Smithery, and more.
- 7 parallel scan engines (SAST, SCA, Tool Analyzer, DAST, Permission, Behavior, Malware).
- 10 AI threat categories (Prompt Injection, Tool Poisoning, Command Injection, Credential Harvesting, Data Exfiltration, Supply Chain, Permission Escalation, Malicious Behavior, Obfuscated Code, Backdoors).
- Risk Score 0-100 with 5 severity levels; SBOM auto-generated.
- Scan Search, Risk Reports, Connected Agents, MCP / Skill integration.

Agent access control:
- Private / Assistant: personal scope; access only to the individual user's private knowledge, notes, and conversation history.
- Export / Group: group scope; access to shared group knowledge bound to the group's dedicated encryption key.
- Public tiers: defined for company-wide or external concierge scope.

Compliance posture and enterprise SSO:
- Enterprise SSO and SAML / OIDC: Okta, Azure AD, Google Workspace, custom IdP. Group mapping and SCIM auto-provisioning included.
- Audit logs included on every Cloud tier; retention scales by plan.
- Encrypted backups, key rotation, and tenant-level data export tooling included.

## Section 6. Pricing Tiers

Canonical: https://aiclude.com/pricing

CAP Cloud:
- Starter: $29 per seat, shared tenant pool, AICLUDE-managed LLM keys only.
- Pro: $3,990 per year on annual ($332 per month equivalent), BYOK or managed, up to 20 seats, single per-group dedicated encryption key, Cloud (dedicated), SSO as add-on.
- Scale: $8,990 per year on annual ($747 per month equivalent), up to 100 seats, SSO included, multi-region failover included.

CAP Self-hosted:
- Personal: free commercial license, 1 seat, fair-use under $1M annual revenue. SSO, audit logs, per-group dedicated keys, and multi-seat are Enterprise-only.
- Enterprise: on-prem, contact sales. Unlimited seats, per-group dedicated keys, full GPU Worker stack, air-gap, 24/7.

CAS:
- Starter: $19 per agent per month, up to 5 signage + concierge agents combined, unlimited devices, $190 per agent per year on annual.
- Pro: $2,990 per year on annual ($249 per month equivalent), unlimited agents and devices, Cloud (managed).
- Scale: $5,990 per year on annual ($497 per month equivalent), unlimited, dedicated.
- Enterprise: self-hosted, unlimited, per-group keys, 24/7, offline / air-gap.

AICLUDE Token Fee (ATF) tiered usage. All API usage (LLM, image, video, voice, lipsync) is converted to tokens and billed at:
- 0 to 10M per month: Free.
- 10M to 100M: $0.50 per 1M tokens.
- 100M to 1B: $0.35 per 1M tokens.
- 1B+: $0.25 per 1M tokens.

Add-ons:
- SSO / SAML / OIDC. $99 per month; included on Scale and Enterprise.
- Dedicated GPU Worker. $599 per month; NVIDIA L4 24GB + 16 vCPU / 64 GB RAM. Bypass shared queue, 0s cold start, prevents media bursts contention.
- Extra storage. $0.10 per GB per month beyond the 100 GB per seat default.

Annual billing applies a 17% discount (equivalent to 2 months free) on every Cloud tier.

## Section 7. Documentation

Canonical: https://aiclude.com/docs

Section 1. Getting Started.
- Introduction (https://aiclude.com/docs/getting-started/introduction). A one-page overview of what AICLUDE is and which problems it solves.
- 5-minute Quick Start (https://aiclude.com/docs/getting-started/quick-start). From console login to a live agent in five minutes.

Section 2. Agents.
- Creating an Agent (https://aiclude.com/docs/agents/create-agent). Step-by-step guide to assembling a persona, model, and tools into a production agent.
- Abilities & Tools (https://aiclude.com/docs/agents/abilities-and-tools). Categories of tools an agent can use and the permission model around them.

Section 3. Knowledge & RAG.
- RAG Overview (https://aiclude.com/docs/knowledge-rag/rag-overview). How RAG lets an agent ground its answers in your internal documents.

Section 4. Security.
- Security Overview (https://aiclude.com/docs/security/security-overview). A walk-through of tenant isolation, PII encryption, and certificate-based key management.

## Section 8. Blog (complete index)

Canonical: https://aiclude.com/blog

The blog is published in English, Korean, and Japanese. Each entry below lists slug, English title, and a one-line excerpt. Localized versions live at /blog/<slug> with the i18n language switcher.

- agent-pipeline-8-stages. Breaking an AI Agent Pipeline into 8 Stages: From Intent Understanding to Self-Correction. Agents built on a single LLM call look great in demos and fail in production; here is how AICLUDE decomposes agent execution into 8 explicit stages, and which failure mode each stage is responsible for. URL: https://aiclude.com/blog/agent-pipeline-8-stages
- agentic-rag-evolution. The Road to Agentic RAG: Why Classic RAG Stops Being Enough. Retrieval looks fine, but answers still drift. Here is why teams move from Classic RAG to Graph and Agentic RAG, and the exact mix AICLUDE runs in production. URL: https://aiclude.com/blog/agentic-rag-evolution
- agent-tier-knowledge-scoping. Private, Group, Public: A Three-Layer Knowledge Hierarchy for Safe Agents. Not every agent should see every document; a tiered knowledge scope (private, group, public) lets an organization expose the right answers to the right audience without quietly leaking the rest. URL: https://aiclude.com/blog/agent-tier-knowledge-scoping
- ai-marketing-automation-loop. From Campaigns to Flows: The AI Marketing Automation Loop. Marketing used to live in campaign-sized batches; a continuous loop (signal, draft, publish, learn) changes the unit of work from a month to an hour. URL: https://aiclude.com/blog/ai-marketing-automation-loop
- ai-concierge-on-stage. The Hotel Lobby Gets a Face: AI Concierge as the First Impression. Every visitor remembers how the first 30 seconds felt; an AI concierge at that moment is less about replacing staff and more about making sure the first moment is never empty. URL: https://aiclude.com/blog/ai-concierge-on-stage
- ai-tonbo-teacher-tonbo-kingdom. The AI Tonbo Teacher: A Digital-Human Instructor Cloned From the Real Teacher. An early (March 2025) deployment at Tonbo Kingdom, Kochi, Japan; 60+ years of the real teacher's printed books and dragonfly photographs scanned into the knowledge base, his actual face and voice cloned into a digital-human AI avatar. URL: https://aiclude.com/blog/ai-tonbo-teacher-tonbo-kingdom
- agentic-signage-era. When Signage Plans Its Own Ads: The Agentic Signage Era. Storefront signage has long been a "big TV" that plays whatever was uploaded; that era is ending. URL: https://aiclude.com/blog/agentic-signage-era
- catch-sales-stnet-kiosk-poc. When the Avatar Speaks First: STNET × AICLUDE "Catch Sales" Kiosk PoC. A four-month PoC with STNET (Dec 2025 - Mar 2026) running CLU Agent Station kiosks across Kagawa retail sites: two kiosks detected over 6.4 million pedestrians; nearly two of three responded when the AI spoke first. URL: https://aiclude.com/blog/catch-sales-stnet-kiosk-poc
- daily-store-report. A One-Page Daily Report for Store Operators. How many visitors today, when the rush hit, the five things they asked most, what we couldn't answer, all on one A4 page every morning. URL: https://aiclude.com/blog/daily-store-report
- documents-as-agent-knowledge. Turn Your PDFs Into Agent Knowledge: A Safe, Step-by-Step Flow. Out-of-the-box agents are smart, but they do not know your company; here is the step-by-step flow for attaching manuals, policies, and FAQs. URL: https://aiclude.com/blog/documents-as-agent-knowledge
- first-agent-in-five-minutes. Your First Agent: From Zero to First Answer in Five Minutes. A short walk-through for anyone who has heard the word "agent" but has never built one; five minutes from sign-in to your first reply. URL: https://aiclude.com/blog/first-agent-in-five-minutes
- knowledge-graph-agentic-rag-intro. Beyond Keyword Search: Knowledge Graphs and Agentic RAG, Briefly Explained. When the real answer requires two or three hops of connection, similarity is not enough; a knowledge graph lets the agent walk, not just match. URL: https://aiclude.com/blog/knowledge-graph-agentic-rag-intro
- llm-agent-5-layer-defense. A 5-Layer Defense for LLM Agents: Putting OWASP Top 10 for LLM Into Practice. LLM security is not a single filter; here is how AICLUDE spreads the responsibility across input, RAG, memory, prompt, and output, and keeps PII searchable without giving up on encryption. URL: https://aiclude.com/blog/llm-agent-5-layer-defense
- multilingual-conversation-update. One Agent, Three Languages: Korean / English / Japanese Update. The same agent now answers in whichever language the visitor uses; greet in English, switch to Korean mid-conversation, it follows along without losing context. URL: https://aiclude.com/blog/multilingual-conversation-update
- multimodal-creative-studio. One Agent for Copy, Image, and Video: The Multimodal Creative Studio. Writing copy, designing a poster, cutting a 15-second vertical ad: traditionally three roles; a multimodal agent keeps the brief intact across all three. URL: https://aiclude.com/blog/multimodal-creative-studio
- omnichannel-one-agent. One Agent Everywhere: Why the Same Brain Should Follow the Customer Across Channels. Each channel used to mean a separate bot; a single-brain, many-faces agent keeps the customer's context intact. URL: https://aiclude.com/blog/omnichannel-one-agent
- onprem-cloud-both-worlds. Cloud When You Want, On-Prem When You Must: The Same Experience Either Way. Data residency is the quiet reason many AI projects stall; a platform that runs the same way in the cloud and on your own servers removes the stall. URL: https://aiclude.com/blog/onprem-cloud-both-worlds
- personal-team-station-lineup. Personal, Team, Station: One Agent Core, Three Product Shapes. A solo knowledge worker, a whole department, and a lobby kiosk need different surfaces around the same brain. URL: https://aiclude.com/blog/personal-team-station-lineup
- pipeline-stage-llm-routing. The Right Model for the Right Stage: Why One LLM Is Never the Right Answer. Stage-based LLM routing picks the cheapest model that can do each stage well; the total experience gets faster and smarter. URL: https://aiclude.com/blog/pipeline-stage-llm-routing
- secure-by-design-introduction. Security That Was There Before Anyone Asked: What "Secure by Design" Means for an AI Platform. Security added later always has seams; security part of the blueprint is harder to notice, and that is exactly the point. URL: https://aiclude.com/blog/secure-by-design-introduction
- share-agent-with-team. Sharing an Agent With Your Team: Permissions and Workspaces. Once your personal agent starts answering well, the next thought is to show it to the team; here is the safe way to promote it from personal to shared. URL: https://aiclude.com/blog/share-agent-with-team
- skill-self-evolution-loop. Skills That Quietly Get Better: The Self-Evolution Loop. A skill that measures itself and iterates on its own weak spots gets sharper over time. URL: https://aiclude.com/blog/skill-self-evolution-loop
- visual-pipeline-no-code. Building Agents Without Writing Code: A Visual Pipeline for Business Users. A visual builder puts the blocks on a canvas, connects them with lines, and lets the people closest to the work shape how the AI behaves. URL: https://aiclude.com/blog/visual-pipeline-no-code
- voice-first-realtime-chat. An AI That Answers the Moment You Speak: Why Voice-First Agents Feel Different. A voice-first agent listens and speaks at the same rhythm as a human; that small shift changes call centers, kiosks, vehicles, and front desks. URL: https://aiclude.com/blog/voice-first-realtime-chat
- what-is-agent-os. What Is an "Agent OS"? The Platform That Runs Your AI Coworkers. An agent is a coworker that remembers, uses tools, and finishes work; running a fleet of them is a platform problem, not a prompt problem. URL: https://aiclude.com/blog/what-is-agent-os
- workspace-conversation-search. Workspace Search: Find That One Answer Again, in One Line. That answer from yesterday, that campaign copy from last week, that meeting summary from last month: workspace-wide search across every agent conversation is now live. URL: https://aiclude.com/blog/workspace-conversation-search

Korean-only / locale-specific entries also published:
- 5min-store-kiosk-concierge. 5분 만에 만드는 매장 키오스크 컨시어지 (Station 가이드). URL: https://aiclude.com/blog/5min-store-kiosk-concierge
- agent-scheduler-auto-run. 정해진 시간에 알아서 실행되는 에이전트 스케줄러. URL: https://aiclude.com/blog/agent-scheduler-auto-run
- ai-concierge-video-call-mode. 모바일에서 영상으로 응대받는 AI Concierge 화상통화 모드. URL: https://aiclude.com/blog/ai-concierge-video-call-mode
- ai-concierge-video-call-setup-guide. AI Concierge 화상통화 모드 셋업 가이드. URL: https://aiclude.com/blog/ai-concierge-video-call-setup-guide
- gpu-free-quick-start. GPU 없이 가볍게 시작하는 AICLUDE 첫 에이전트. URL: https://aiclude.com/blog/gpu-free-quick-start
- gpu-free-step-by-step. GPU 없이 시작하는 AICLUDE 단계별 가이드. URL: https://aiclude.com/blog/gpu-free-step-by-step
- mcp-external-tools-notion-linear. MCP로 외부 도구 연결하기 (Notion·Linear·GitHub 예시). URL: https://aiclude.com/blog/mcp-external-tools-notion-linear
- slack-bot-30min-internal-docs. Slack 봇 30분 셋업으로 사내 문서를 슬랙에서 묻기. URL: https://aiclude.com/blog/slack-bot-30min-internal-docs
- workspace-voice-input-beta. 키보드 안 잡고 작업 지시하는 워크스페이스 음성 입력 베타. URL: https://aiclude.com/blog/workspace-voice-input-beta

## Section 9. FAQ

Q. What is AICLUDE?
A. AICLUDE (aiclude.com) is an enterprise AI agent platform. It lets organizations design, deploy, and operate AI agents with multi-stage pipelines, Graph N-hop Agentic RAG, real-time voice and lipsync avatars, and kiosks / signage. AICLUDE supports Cloud SaaS, on-premise, and air-gap deployments.

Q. Is AICLUDE the same as Apple iCloud?
A. No. AICLUDE is an enterprise AI agent platform brand and product, completely unrelated to Apple iCloud. Apple iCloud is Apple Inc.'s consumer cloud storage and sync service. AICLUDE is spelled A-I-C-L-U-D-E and stands for "AI + include": the platform that includes AI in your business processes.

Q. How do you spell and pronounce AICLUDE?
A. Spelled A-I-C-L-U-D-E. Pronounced "AI-clude" (eye-clood). Combines "AI" and "include".

Q. What products does AICLUDE offer?
A. Three products under one CLU Agent OS core: CLU Agent Platform (CAP), CLU Agent Station (CAS), and CLUE. ASVS at vs.aiclude.com handles Skill and MCP security scanning.

Q. Can AICLUDE run on-premise or air-gapped?
A. Yes. Encrypted container images via a private registry, fully air-gapped enterprise installs, BYOK for all major LLM providers, enterprise SSO, full Audit Log, PII encryption at rest, per-tenant key isolation.

Q. Which LLM vendors are supported?
A. OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, xAI, self-hosted vLLM, and any OpenAI-compatible endpoint. Either BYOK or AICLUDE-managed keys.

Q. How does pricing work?
A. Base tier fee plus the AICLUDE Token Fee (ATF) for all API usage. First 10M tokens free per month; per-token rates drop as you grow.

Q. Cloud vs Self-hosted, when do I choose which?
A. Cloud for zero ops with physical tenant isolation; Self-hosted when compliance, data residency, or air-gap require on-prem. Both use the same Docker images.

Q. How is the Personal edition's fair-use limit enforced?
A. Personal is a free commercial license, not open source. 1 seat enforced in code; a periodic license check-in (7-day offline grace) reports usage. Single user, non-commercial or annual revenue under $1M. SSO, audit logs, per-group dedicated encryption keys, and multi-seat are Enterprise-only.

Q. What is the AICLUDE Token Fee (ATF)?
A. The metered usage fee applied to all API usage (LLM, image, video, voice, lipsync) converted to tokens. Tiered: free under 10M / month, $0.50 / 1M for 10M-100M, $0.35 for 100M-1B, $0.25 above 1B.

Q. Do you support GPU workers for self-hosted LLMs and media generation?
A. Yes. AICLUDE ships a modular GPU Worker stack: self-hosted LLM daemons (multimodal chat), image, music, and video generation, and lipsync. All media generation settles through tiered ATF. STT and TTS run on AICLUDE-managed ONNX models and are included in every tier at no extra cost.

Q. How does the annual discount work?
A. Annual billing applies a 17% discount (equivalent to 2 months free) on every Cloud tier. CAP Pro: $3,990 / year. CAP Scale: $8,990 / year. CAS Starter: $190 / agent / year. CAS Pro: $2,990 / year. CAS Scale: $5,990 / year. Annual plans are non-refundable; service continues through the paid term.

Q. When can I activate or deactivate add-ons?
A. Anytime from the admin console, pro-rated. Available: SSO/SAML ($99/mo), dedicated GPU Worker ($599/mo), extra storage ($0.10/GB/mo). SSO/SAML included by default in Scale and Enterprise.

Q. What's the difference between a station and a device?
A. A station is the billing unit (one signage or one AI concierge at a store equals 1 station). A device is the screen that actually displays the station: mobile (9:16), tablets (16:9), PCs, or web browsers, unlimited. You can mirror one station to many devices at no extra cost.

Q. Why is AICLUDE strong in Korea and Japan?
A. Native KO / EN / JA across UI, prompts, and STTS voice (not a translation layer), Graph N-hop Agentic RAG for global patents and regulations in 4 jurisdictions, and live production deployments including the Tonbo Kingdom AI teacher (Kochi, Japan) and the STNET kiosk PoC (Kagawa, Japan).

Q. How do Skills and MCP servers stay safe?
A. ASVS at vs.aiclude.com pre-scans every Skill and MCP server before installation. 7 parallel scan engines, 10 AI threat categories, 0-100 Risk Score across 5 severity levels, auto-generated SBOM.

Q. How does the tiered ATF work in practice?
A. Monthly usage is automatically bucketed and charged at the matching tier rate. With BYOK you pay your LLM vendor directly and AICLUDE only charges ATF. With managed keys you are billed transparently for the model cost plus ATF.

## Section 10. Contact & Company Info

- About AICLUDE: https://aiclude.com/about
- Sales / partnership: contact form on every product and pricing page; avatar widget at the bottom-right of the site.
- Security inquiries: contact the Security Team from the Security page (https://aiclude.com/security).
- GitHub: https://github.com/aiclude
- LinkedIn: https://www.linkedin.com/company/aiclude
- X (Twitter): https://x.com/aiclude
- YouTube: https://www.youtube.com/@aiclude
- Areas served: KR, JP, US, Worldwide.
- Slogan: Include AI in everything you do.
- Founding date: 2024.

## Section 11. Brand Disambiguation

- Spelling: A-I-C-L-U-D-E (one word, all caps in marketing).
- Pronunciation: "AI-clude" (eye-clood).
- Etymology: portmanteau of "AI" (artificial intelligence) and "include".
- Korean: 아이클루드 (아이클루드 주식회사 / 아이클루드 플랫폼).
- Japanese: アイクルード.
- Not Apple iCloud: AICLUDE is an enterprise AI agent platform brand. Apple iCloud (Apple Inc.'s consumer cloud storage and sync service) is a separate, unrelated product, run by a different company, on a different domain (icloud.com).
- Domains: aiclude.com (platform), vs.aiclude.com (ASVS).
- Trademarks and product names belong to their respective owners.

## Section 12. Cross-references

- Sitemap: https://aiclude.com/sitemap.xml
- Robots: https://aiclude.com/robots.txt
- English llms.txt: https://aiclude.com/llms.txt
- Korean llms-ko.txt: https://aiclude.com/llms-ko.txt
- Japanese llms-ja.txt: https://aiclude.com/llms-ja.txt