About iPassion AI Portal¶

What is iPassion AI Portal?¶

iPassion AI Portal is an Enterprise AI Platform by iPassion Co., Ltd., designed to serve as a centralized hub for AI usage within your organization. It combines Generative AI, Knowledge Management, and AI resource management capabilities into a single platform.

The system is built on Open WebUI, a highly popular open-source AI platform, enhanced with additional features to meet enterprise requirements.

Key Features¶

Chat and Conversations¶

Multi-model chats -- select one or more models per conversation and compare responses side by side.
File and image uploads -- attach documents, images, and other files directly in the chat for AI analysis.
Web search -- enable real-time web search to ground responses in up-to-date information.
Code execution -- run code snippets (Python and more) directly within the chat via the built-in code interpreter.
Message queue -- messages are queued and processed reliably, even under heavy load.
Memory -- the system can remember facts and preferences across conversations for a personalized experience.
Folders, tags, and pins -- organize conversations into folders, apply tags, and pin important chats for quick access.
Voice and audio -- speech-to-text input and text-to-speech playback; supports configurable TTS/STT engines.
Image generation -- generate images from text prompts using integrated image generation backends (DALL-E, ComfyUI, Automatic1111, etc.).
Automations -- define event-driven automations that trigger model actions on specific conditions (e.g., @-mention a model in a channel).

Knowledge and RAG (Retrieval-Augmented Generation)¶

Multiple vector database backends -- PGVector (used in production), ChromaDB, Milvus, Qdrant, OpenSearch, Elasticsearch, and more.
Hybrid search -- combines BM25 keyword search with vector similarity search and cross-encoder reranking for high-accuracy retrieval.
Multiple content extraction engines -- extract text from PDFs, Word, PowerPoint, Excel, HTML, and other formats using engines such as Tika, Docling, and others.
Agentic retrieval -- the model autonomously decides when and how to query knowledge collections during a conversation.
Full context mode -- inject the entire content of a knowledge collection directly into the prompt when the document set is small enough.
Multiple file formats -- PDF, DOCX, PPTX, XLSX, TXT, JSON, HTML, Markdown, CSV, and more.

Models and Agents¶

Model presets -- create custom model profiles with system prompts, default parameters (temperature, top-p, etc.), and bound tools or knowledge collections.
Dynamic variables -- use template variables in system prompts such as {{USER_NAME}}, {{CURRENT_DATE}}, {{USER_LOCATION}}, and {{USER_LANGUAGE}} for personalized instructions.
Bound tools and knowledge -- attach specific tools and knowledge collections to a model so they are always available when that model is selected.
Access control -- restrict which users or groups can access specific models or agents.
Global model defaults -- administrators can set organization-wide default parameters for all models.

Notes¶

Rich text editor -- create and organize personal or shared notes with a full rich-text editor.
AI Enhance -- highlight text within a note and use AI to rewrite, summarize, expand, or translate it.
Context injection -- reference notes directly in chat conversations to provide the model with additional context.

Channels¶

Team collaboration spaces -- create topic-based channels for group discussions, similar to team messaging platforms.
@model tagging -- mention an AI model (e.g., @gemini-2.5-flash) inside a channel message to get an AI response visible to all members.
Threads and reactions -- reply in threads to keep conversations organized; use reactions for quick feedback.

Extensibility¶

Tools (Python) -- write custom Python tools that extend model capabilities directly within the UI (managed via Workspace > Tools).
Skills -- reusable prompt/instruction packs that bundle system prompts and configurations for specific use cases (managed via Workspace > Skills).
Pipelines -- build multi-step processing pipelines for advanced workflows (input/output filters, RAG pipelines, function-calling pipelines).
MCP (Model Context Protocol) support -- connect to external MCP servers to provide models with additional context and tool access.
OpenAPI servers -- register any OpenAPI-compatible server as a tool source; the model can call its endpoints automatically.
Prompts library -- save, share, and manage reusable prompt templates with variable placeholders (managed via Workspace > Prompts).

Authentication and Access¶

Role-Based Access Control (RBAC) -- permissions for Admin, Group Admin (ผู้ดูแลแผนก), User, and Pending roles.
SSO / OIDC / LDAP -- integrate with enterprise identity providers via OpenID Connect, OAuth 2.0, or LDAP. Configurable from the Admin Panel (26 settings fields).
SCIM 2.0 -- automated user provisioning and de-provisioning from identity management systems.
API keys -- generate personal API keys for programmatic access to the platform.

Administration¶

Analytics dashboard -- view token consumption and model usage breakdown in the Admin Panel.
Model evaluation and arena -- run blind model comparisons (arena mode) and structured evaluations to benchmark model quality.
Banners -- display system-wide announcements and maintenance notices to all users.
Webhooks -- send event notifications to external systems (e.g., new user registration, chat events).

Platform-Specific Additions¶

System Components¶

iPassion AI Portal consists of 2 main items:

#	Item	Description
1	Enterprise AI Platform (AI Portal)	Web application for organization-wide AI usage -- 1 system, 3-year license
2	AI Token Usage Credit	Token credits for AI model usage -- 1 set, covering 3 years

User Roles¶

Role	Description	Key Permissions
Admin (ผู้ดูแลระบบ)	System administrator	Manage users, groups, models, connections, system settings
Group Admin (ผู้ดูแลแผนก)	Department administrator	Manage group members, view group usage reports, manage group Knowledge Bases
User (ผู้ใช้)	General user	Chat with AI, use Knowledge Bases, use AI Agents, manage personal conversations
Pending (รอการอนุมัติ)	Awaiting approval	Cannot access the system until approved by Admin

Guardrails (PII Detection)¶

PII Guardrail -- a built-in filter function that detects and blocks personally identifiable information (phone numbers, email addresses, Thai national ID numbers) in AI responses using regex and checksum validation.
Additional guardrails can be configured via System Prompts at the system or agent level.

Multi-LLM Provider Support¶

The system supports connecting to multiple AI providers simultaneously via the OpenAI-compatible API. Administrators add connections through Admin Panel > Settings > Connections by entering each provider's API Base URL and API Key. Users simply select from the available models -- no API key management required on the user side.

Supported providers include Google Gemini, OpenAI, OpenRouter (unified gateway to Claude, Mistral, and hundreds of other models), Ollama (local models), and any OpenAI-compatible API.

Info

Anthropic Claude is accessible via OpenRouter rather than direct connection, as Anthropic uses a non-standard authentication header.

Token Usage Tracking

The system tracks actual token usage (input and output separately) per user and per model. Users can view their consumption via Settings > Usage. Administrators can view system-wide analytics via Admin Panel > Dashboard.

System Requirements¶

Client Requirements¶

Web Browser: Google Chrome 90+, Microsoft Edge 90+, Mozilla Firefox 90+, Safari 14+
No additional software installation required on client machines
Supports desktop, tablet, and mobile devices

Server Requirements¶

Operating System: Linux (Ubuntu 22.04+ / RHEL 8+)
Container Runtime: Docker Engine 24.0+ / Docker Compose 2.20+
RAM: 8 GB minimum (16 GB recommended)
Storage: 50 GB minimum for system + Knowledge Base data
Database: PostgreSQL 15+ with pgvector extension
Internet connectivity for accessing AI provider APIs