iKB by ThoughtsMachine

Enterprise-grade AI knowledgebase with governed accuracy

5 followers

Enterprise-grade AI knowledgebase with governed accuracy

5 followers

Visit website

Knowledge base software

•

AI Chatbots

•

Business intelligence software

iKB is a self-hosted AI knowledge base platform that enables organisations to build conversational interfaces over their document libraries. It combines vector-based retrieval with automatic knowledge graph construction, delivering significantly higher accuracy on complex queries requiring multi-document synthesis.

Interactive

Free Options

Launch tags:SaaS•Artificial Intelligence•Bots

Launch Team / Built With

Figr AI: UX Agent for Product Teams — Learns your product. Thinks through UX

Learns your product. Thinks through UX

Promoted

Maker

📌

Hi Product Hunt, I’m Sanjay Willie from ThoughtsMachine. Over the past few months, I’ve been building iKB an AI powered knowledge base platform for teams who need more than basic RAG. I kept looking for something like this in the market, but most options either felt designed for personal use, or they didn’t have the knowledge sharing capabilities I needed. So I built one. It started as a way for us to share our own technical knowledge base with customers, but it quickly evolved into a platform our clients asked to use for their own internal knowledge as well. What is iKB? iKB is a self hosted platform that lets you build conversational interfaces over your document libraries. It combines traditional vector based retrieval with knowledge graph construction, which materially improves accuracy for complex, multi document queries. Users can create topics, include one or many documents in those topics, including dimensional data like worksheets (excel, etc) and can perform analytics from it by dynamically generating isolated pandas queries. It can also ingest open drawing format (DXF), extract layers and visually understand drawings like humans could. 1) Trust, Safety, and “No Leaks” If iKB can’t be trusted with sensitive internal knowledge, nothing else matters. The first priority is making sure confidential content stays confidential, and that we can confidently deploy it without fear of data exposure. 2) Control Over Who Sees What In a real organisation, not everyone should see everything. The next most important value is the ability to define access cleanly by department, team, group, topic, and channel so the knowledge is shared with the right people and hidden from the wrong ones. 3) Governance and Accountability (Audit Trail) When something changes, or a mistake happens, you need traceability. Business leaders care that there is an audit trail: who did what, when, and from where so the platform supports internal governance, compliance expectations, and operational discipline. 4) Reliability and Operational Stability A business user doesn’t want a “smart demo.” They want something that works every day. The platform needs to be stable, recover cleanly from issues, and handle failures gracefully without breaking workflows or silently losing data. 5) Accurate Answers on Real World Questions (Not Just Simple RAG) This is the point of iKB: accurate answers to complex questions that span multiple documents, policies, departments, and versions. Business users care less about how retrieval works, and more about: “Does it give me the right answer, consistently, with the right sources?” 6) Proof of Answer Quality (Measurable Confidence) Beyond “it seems correct,” iKB gives a way to measure and monitor quality over time. For a business user, this means you can spot weak areas, track improvement, and build confidence internally especially when the platform becomes relied upon for real decisions. 7) Rollout Across Real Channels (Where Work Actually Happens) A platform is only useful if it fits the day to day workflow. Business users care that iKB can be deployed where employees and customers already communicate web chat, embedded widgets, and omni channels like WhatsApp, Telegram, SMS, email without reinventing the way teams operate. 8) Human Handover When It Matters AI shouldn’t be a dead end. When the user needs help, escalation must be seamless. Business users value clear handover triggers, proper routing, and a clean way for an agent to take over without losing context or creating chaos. 9) Easy Knowledge Ingestion From “Messy Reality” Organisations don’t have perfect documents. They have PDFs, Word files, spreadsheets, slide decks, web pages, and images stored across drives and systems. Business users care that iKB can absorb this reality, keep it organised, and stay updated when content changes. 10) Data Retention Controls and Sensitive Conversation Options Some conversations should be stored, others should not. Business users value the ability to apply retention rules and enable private/ephemeral modes when needed so the platform can be used even in sensitive or regulated contexts. 11) Cost Visibility and Usage Control Once AI is deployed broadly, cost becomes a business problem. Business users want visibility: usage by topic, by channel, by time period, and controls that prevent runaway spend without requiring a data team to interpret it. 12) A Platform That Fits Enterprise Procurement Reality This is often the deciding factor: the company wants ownership and control. Self hosting, data ownership, no lock in, and predictable licensing matter because they reduce procurement friction and long term risk. 13) Administration That Doesn’t Become a Full Time Job Business adoption fails when the admin overhead is too high. Business users care that topics, widgets, channels, users, access rules, and configurations can be managed cleanly preferably with bulk actions and predictable defaults. 14) Branding, UX Polish, and Localization This matters for adoption and user comfort, especially in external facing deployments. But it’s not what makes the platform enterprise grade. It’s what makes it feel natural once the fundamentals above are solid.

Report

30d ago

Maker

Platform Overview

Core capabilities

RAG-based chat with streaming responses and source citations
Hybrid search (vector + keyword) with reranking and deduplication controls
Knowledge graph enhancement (LightRAG/GraphRAG) for multi-hop reasoning
Ingestion for PDFs, Office files, images (OCR), and text formats
Web crawling using Playwright with SSRF protections
Cloud ingestion via rclone and S3-compatible storage (S3/MinIO/R2)
Multi-topic architecture with granular access control and custom settings
Analytics: token usage, session metrics, feedback, ingestion/crawl performance
Enterprise security: encryption, rate limiting, CSRF protection, audit logs

Architecture and Stack

High-level components

Web UI (Admin + Chat), API layer, retrieval engine, citation assembly
Knowledge graph layer (LightRAG/GraphRAG) with per-topic enablement
Data layer: PostgreSQL + pgvector; Redis for caching and rate limiting
Integrations: object storage, rclone cloud drives, Chatwoot, custom AI endpoints

Retrieval modes

Vector search, hybrid (BM25-like + vector), reranking, diversity caps, deduplication

Model management

Per-topic model configuration (OpenAI and OpenAI-compatible endpoints)
Token and pricing tracking (admin-configurable), temperature and response controls

Ingestion and Knowledge Management

Inputs: PDF, DOCX, PPTX, XLSX, images (OCR), TXT/CSV/Markdown

Channels: UI upload, web crawl, S3-compatible storage, rclone cloud drives

Pipeline: validate → (optional) malware scan → extract/OCR → chunk/tokenize → embed → store in pgvector → optional GraphRAG indexing

Security, Compliance, and Governance

Security controls

Encrypted message storage (AES-256-GCM for chat content)
Secure sessions (strict cookies) and CSRF protection
Rate limiting (platform and endpoint levels)
Admin audit logs and security headers (CORS/CSP)
SSRF-safe crawling and IP allowlisting for sensitive callbacks

Governance

Topic-level access control (public/private/unlisted), user groups, topic groups
Admin-only configuration with auditability
Optional incognito mode for sensitive queries

Deployment, Performance, Operations

Deployment: self-hosted, private cloud, on-premises, air-gapped

Reference concurrency: ~30–80 concurrent chats per instance (configuration-dependent)

Scaling levers: horizontal app nodes/workers, PostgreSQL tuning/pooling, batching, dedicated storage for ingestion/graph

Operations: background tasks for indexing, health checks, logging, query timeouts, configurable upload/crawl limits

Differentiators

Accuracy-first design: citations, hybrid retrieval, and governance-first controls
GraphRAG/LightRAG augmentation for deeper, multi-hop reasoning
Flexible deployment including air-gapped
Broad integrations (Chatwoot OMNI, rclone, S3-compatible storage)

Report

29d ago

Hey Sanjay,
Quick nudge. I revisited your page and the idea is strong, but the value of faster knowledge resolution isn’t fully surfaced upfront yet.
I’ve got some specific thoughts on tightening that narrative if you want.

Report

15d ago

Platform Overview

Core capabilities

RAG-based chat with streaming responses and source citations
Hybrid search (vector + keyword) with reranking and deduplication controls
Knowledge graph enhancement (LightRAG/GraphRAG) for multi-hop reasoning
Ingestion for PDFs, Office files, images (OCR), and text formats
Web crawling using Playwright with SSRF protections
Cloud ingestion via rclone and S3-compatible storage (S3/MinIO/R2)
Multi-topic architecture with granular access control and custom settings
Analytics: token usage, session metrics, feedback, ingestion/crawl performance
Enterprise security: encryption, rate limiting, CSRF protection, audit logs

Architecture and Stack

High-level components

Web UI (Admin + Chat), API layer, retrieval engine, citation assembly
Knowledge graph layer (LightRAG/GraphRAG) with per-topic enablement
Data layer: PostgreSQL + pgvector; Redis for caching and rate limiting
Integrations: object storage, rclone cloud drives, Chatwoot, custom AI endpoints

Retrieval modes

Vector search, hybrid (BM25-like + vector), reranking, diversity caps, deduplication

Model management

Per-topic model configuration (OpenAI and OpenAI-compatible endpoints)
Token and pricing tracking (admin-configurable), temperature and response controls

Ingestion and Knowledge Management

Inputs: PDF, DOCX, PPTX, XLSX, images (OCR), TXT/CSV/Markdown

Channels: UI upload, web crawl, S3-compatible storage, rclone cloud drives

Pipeline: validate → (optional) malware scan → extract/OCR → chunk/tokenize → embed → store in pgvector → optional GraphRAG indexing

Security, Compliance, and Governance

Security controls

Encrypted message storage (AES-256-GCM for chat content)
Secure sessions (strict cookies) and CSRF protection
Rate limiting (platform and endpoint levels)
Admin audit logs and security headers (CORS/CSP)
SSRF-safe crawling and IP allowlisting for sensitive callbacks

Governance

Topic-level access control (public/private/unlisted), user groups, topic groups
Admin-only configuration with auditability
Optional incognito mode for sensitive queries

Deployment, Performance, Operations

Deployment: self-hosted, private cloud, on-premises, air-gapped

Reference concurrency: ~30–80 concurrent chats per instance (configuration-dependent)

Scaling levers: horizontal app nodes/workers, PostgreSQL tuning/pooling, batching, dedicated storage for ingestion/graph

Operations: background tasks for indexing, health checks, logging, query timeouts, configurable upload/crawl limits

Differentiators

Accuracy-first design: citations, hybrid retrieval, and governance-first controls
GraphRAG/LightRAG augmentation for deeper, multi-hop reasoning
Flexible deployment including air-gapped
Broad integrations (Chatwoot OMNI, rclone, S3-compatible storage)