Governed AI Knowledge Platform | Four-Pillar Audit System

How It Works

From Documents to Intelligent Answers

1

Upload Your Knowledge

Import documents, manuals, FAQs, and any text-based content. iKB processes and indexes everything automatically.

2

AI Understands Context

Advanced NLP extracts meaning, relationships, and context from your content.

3

Users Ask Questions

Customers or employees ask questions in natural language across 6 channels.

4

Accurate, Sourced Answers

iKB retrieves relevant information and generates accurate, cited responses.

Governance & Compliance

Enterprise-Grade AI Governance

Four independent, immutable logging systems and a dedicated superadmin dashboard. Every action captured. Nothing deleted. Full non-repudiation.

Four-Pillar Audit System

Audit Logs

Complete record of all admin, user, and system actions with actor, IP, user-agent, HTTP method, and status code.

Prompt Versions

Every AI instruction change tracked with auto-incrementing version numbers per entity+field pair. Full old/new value history with side-by-side comparison.

Config Changes

47+ topic configuration fields tracked with before/after snapshots. Sensitive values (API keys, passwords) automatically redacted in logs.

Moderation Events

Every content moderation API call logged with result, response time, flagged categories, and summary statistics across 13 content categories.

Governance Design Principles

Immutability — All records append-only, deletion returns 403 Forbidden
Fail-Safe — Logging failures never break main operations
Separate Transactions — Governance commits independently after main operation
Race Condition Prevention — SELECT FOR UPDATE SKIP LOCKED for multi-worker safety
Privacy by Design — Sensitive values auto-redacted as [REDACTED]
Non-Repudiation — IP, user-agent, timestamp, actor for every action

Content Moderation

Built-in 13-category content moderation using OpenAI's omni-moderation-latest model, free of charge. Configurable per topic, enforced across all channels.

Fail-Closed Default — API errors block the message (secure default)
Per-Topic Toggle — Enable/disable and select categories per knowledge base
Per-Channel Coverage — Native chat, widget, Teams, Slack, OMNI
Performance Tracking — Response time recorded for every moderation call
User Warnings — Translated moderation messages when content is flagged

Categories: Harassment, Threatening Harassment, Hate Speech, Threatening Hate, Sexual Content, CSAM Detection, Violence, Graphic Violence, Self-Harm, Self-Harm Intent, Self-Harm Instructions, Illicit Content, Illicit Violence.

Security & Encryption

Multiple security layers protect your data at every level. Two comprehensive security audit rounds covering 38 findings across 19+ files.

Dual-Key Encryption — AES-256-GCM with user key + admin key. All API keys and secrets Fernet-encrypted. Sensitive values auto-redacted in governance logs.
No External Training — Your documents are NEVER used to train AI models.
Enterprise SSO — Authentik OIDC with headless auth, JIT provisioning, group sync, admin auto-provisioning. Three modes: SSO-only, hybrid, local-only.
Application Hardening — SSRF blocking, ODBC injection prevention, encryption fail-closed, XSS encoding, CSP nonces, CSRF protection, HSTS.
Session Security — HTTPOnly cookies, strict SameSite, UUID v4 session validation, TOTP-based 2FA, HMAC webhook verification, IP whitelisting.

GDPR & Data Privacy

A dedicated DSAR module under Governance gives superadmins complete GDPR compliance tooling.

Search — Find any user by email, view complete data inventory across all tables
Export (Art. 15) — One-click JSON export of all user data in structured, human-readable format
Purge (Art. 17) — Complete data erasure with confirmation safeguard. Anonymises user account, deletes messages, sessions, and webchat data
Audit-Logged — All DSAR actions recorded in the Governance dashboard
Compliance-Safe — Audit logs preserved even after purge for regulatory compliance

Data inventory includes: Sessions, Messages, Feedback, Access Grants, Group Memberships, Shared Conversations, Webchat Sessions, Audit Logs.

AI Intelligence

Advanced Retrieval & Quality

Multi-stage retrieval pipeline with quality scoring, self-critique, and automatic regeneration. Accuracy demands more than keyword matching.

Retrieval & Quality (RAG)

Knowledge Graph (GraphRAG) — Cross-document relationships, entity extraction, hybrid graph+vector queries with 15+ configurable settings, multi-turn query enhancement.
RAG Strategy Recommender — Analyzes content types and recommends optimal retrieval strategy: vector search, hybrid search, or GraphRAG.
Advanced RAG Pipeline — HyDE retrieval, multi-query variations, agentic retrieval, and LLM-based re-ranking.
Quality Scoring & Self-Critique — Five-dimension evaluation: faithfulness, relevance, groundedness, confidence, and overall composite. Automatic regeneration on low scores.
Multi-Model Support — OpenAI (GPT-4o, GPT-5, o1, o3), Anthropic Claude, Google Gemini, Cohere. Per-topic model assignment.
Multi-Language — Ask in any language, get answers in any language. Default UI in Bahasa Malaysia with full i18n support.
Source Citations — Every answer shows source documents with page numbers. Per-document citation control and direct page links.
Spreadsheet Analytics — Natural language queries on spreadsheets with sandboxed Python execution, multi-sheet JOINs, and result caching.

Knowledge Organization

Organize by Topic — Create separate knowledge bases for different products, teams, or use cases with independent AI settings and documents.
Group Related Topics — Organize topics into category groups for easy navigation. Multi-product companies or complex teams.
Conversation Memory — Follow-up questions naturally. AI remembers context within sessions.
Custom AI Instructions — Set custom instructions per topic. Match your brand voice and control response style.
Suggested Questions — Configure starter questions to guide users toward common queries.
Real-Time Updates — Update documents and see changes reflected immediately.

Topic Configuration

Fine-tune every knowledge base with granular controls.

Starter Forms — Pre-chat forms with three modes: Free Text, Form First (skippable), Form Only (required).
AI Disclosure Text — Configurable disclosure notice below chat input for regulatory transparency. Appears in native chat and webwidget.
Config Summary — Read-only governance summary page per topic: identity, AI model settings, RAG config, Knowledge Graph, moderation rules, access control, usage stats. Printable.
Quick Activate/Deactivate — Toggle topics on/off from the Topics list with a single click. Governance-logged with before/after state.

Channels & Integration

One Knowledge Base, Every Channel

Deploy across 6 channels, each with independent rate limits, session management, and governance coverage.

💬 Native Chat

▭ Web Widget

👥 Teams

💬 Slack

📱 Chatwoot/OMNI

⚡ REST API

Web Chat Widget

Deploy intelligent chat on your website with a simple embed code.

            <script src="https://your-domain/widget.min.js" data-widget-key="YOUR_KEY"></script>
          

Voice input (STT) & voice output (TTS)
Starter forms / pre-chat questionnaires
Human escalation ("Talk to a Human")
AI disclosure text for regulatory compliance
Lead capture with GDPR consent
Light/dark/auto themes, custom colors
3 modes: bubble, sidebar, centered
Domain whitelisting

Channel Details

Native Chat — Full-featured web interface with streaming and citations
Web Widget — 3 modes (bubble, sidebar, centered), lead capture, domain whitelisting
Microsoft Teams — Native bot with threading and OAuth
Slack — Events API, Slack Blocks, file uploads
Chatwoot/OMNI — CRM integration with agent handover
REST API — Stateful/stateless, Bearer auth, rate limiting

Per-Channel Governance

Independent rate limits, prompt version tracking, and token usage per channel. Content moderation enforced across all channels.

Session Traceability

Per-tab isolation with UUID v4, cross-tab continuity, channel attribution, and configurable session expiry. Admin view across all channels.

Human Escalation

When AI cannot adequately help, users can request human assistance directly from the chat interface.

Per-Topic Toggle — Enable or disable escalation on each knowledge base
Email Notifications — Comma-separated list of recipients receive formatted escalation emails
Chat Transcript — Last 20 messages automatically attached to the escalation email
Custom Button Label — Override the default "Talk to a Human" button text
Rate Limited — 3 escalation requests per hour per user
Cross-Channel — Works in native chat and the embeddable webwidget

Chat Export

Turn any conversation into a professionally formatted document. AI intelligently structures your chat content.

Word (DOCX) — Structured reports with headings, tables, and citations
Excel (XLSX) — Data and analytics extracted into structured spreadsheets
PDF — Print-ready formatted documents for compliance and offline records

AI analyzes conversation content and chooses appropriate formatting: tables for data, bullet points for lists, sections for long-form answers. Enable/disable per topic.

External API Tools

Connect your AI assistant to live data sources and business systems with full governance controls.

MCP Connections — Connect to any MCP-compliant server via Streamable HTTP. Automatic tool discovery, health checks, and admin approval workflow.
REST Tools — Define REST API endpoints with templated URLs, configurable HTTP methods, Jinja2 request body templates, and JMESPath response mapping.
Tool Orchestrator — Parallel execution with per-tool rate limiting (Redis-backed), response caching, and configurable timeouts.
Encrypted Auth — Bearer, API Key, Basic Auth, Custom Headers. All credentials Fernet-encrypted.
Execution Audit — Three retention modes: full, redacted, or metadata-only.
Block Mode — Required tools that fail short-circuit the AI response with a user-facing error.
Actionable vs Read-Only — Tools classified as "actionable" require user confirmation before execution.
Response Redaction — Sensitive/PII tool responses automatically redacted in logs.

SQL Functions Browser

Schema-level discovery and management of SQL functions and stored procedures.

Auto-Discovery — Scans databases for functions and procedures with parameter signatures and return types
Admin Annotations — Descriptions, categories, and highlight/hide flags to control AI references
AI Instructions — Connection-level free-text instructions injected into the AI prompt
Query Playground — Test SQL queries directly from the admin interface
Governance-Logged — All function annotation changes tracked in audit logs

Integration APIs & Cloud Sources

REST API — Full API access for programmatic control. Query knowledge bases, manage documents, retrieve analytics.
Cloud Storage Sync — AWS S3, Google Drive, OneDrive, SharePoint, Dropbox, SFTP, WebDAV, Nextcloud, and 40+ more via rclone.
Analytics & Reporting — Real-time metrics with input/output token split, per-channel tracking, cost projections. Governance dashboard with four tabs.
Feedback System — Three modes: disabled, basic (thumbs), detailed (categories + comments). Admin notes and analytics.
Governance Exports — Per-tab CSV or ZIP bundle with system snapshot JSON. Rate-limited, up to 10,000 records.

Deploy & Scale

Flexible Deployment, Any Scale

Deployment Options

Cloud SaaS — Fully managed, automatic updates and scaling
Private Cloud — Dedicated instance in your cloud region
On-Premise — Your infrastructure, air-gapped available

Document Formats

PDF, Word, Text, Markdown, RTF, BibTeX, EPUB
Excel, CSV, ODS with NL queries & sandboxed Python
DWG and DXF (native AutoCAD)
EML, MSG with attachment extraction
Images: PNG, JPG, GIF, BMP, TIFF, WEBP
Web crawling with configurable depth

Magic byte validation, ClamAV scanning, ZIP bomb protection, UUID filename sanitization.

Performance

Query latency < 500ms (p95)
100+ pages/minute ingestion
10,000+ concurrent users
99.9% uptime SLA
Auto-scaling, global CDN
Daily backups, 30-day retention

Governed AI for Critical Knowledge