What is ASK? Multi-tenant RAG platform for document AI explained

Your team has hundreds of PDFs, policies, manuals, and SOPs. Your customers have questions. Generic ChatGPT does not know your content — and you cannot paste confidential documents into a public chatbot every time someone asks.

ASK is Logic Spine's answer to that gap.

ASK is a multi-tenant RAG SaaS platform at ask.logicspine.com. Organisations upload documents, build a searchable knowledge base, and let users chat with an AI that answers from their content — with citations, streaming responses, and proper access control.

Tagline: Upload. Ask. Answer.

This article explains what ASK is, who it is for, how it works, and how it fits alongside Sumikar and UnderKB in the Logic Spine product family.

The problem ASK solves

Knowledge is trapped in documents. Support teams re-type the same answers. Sales engineers hunt through decks. New hires ask seniors questions that are already written down — somewhere.

Teams try to fix this with:

• Shared drives — files exist, but nobody finds the right paragraph at the right time
• Wiki tools — high maintenance; content goes stale
• Generic AI chat — fluent but hallucinates; no grounding in your actual policies
• Custom RAG builds — powerful but expensive to ship and maintain

ASK productises the middle path: RAG (Retrieval-Augmented Generation) as a ready-made platform — multi-tenant, subscription-ready, and built for teams who need document-grounded AI without assembling infrastructure from scratch.

What ASK actually is

ASK is cloud software where:

1. Organisations (tenants) sign up and get an isolated workspace
2. Admins upload PDFs, DOCX, and other documents into a knowledge base
3. The platform chunks, embeds, and indexes content for hybrid search
4. Users chat with an AI that retrieves relevant passages and streams answers with citations
5. End customers can use a lightweight public chat at /t/{your-org}/chat — without seeing your full admin dashboard

It is not a chatbot wrapper around a single API call. It is a full RAG pipeline: ingestion → verification → embedding → hybrid retrieval → streaming LLM → billing and permissions.

Who ASK is for

Audience	Use case
SaaS companies	Product docs, help centre, onboarding knowledge
Professional services	Client playbooks, proposals, compliance packs
Internal ops teams	SOPs, HR policies, IT runbooks
Education & training	Course materials students can query
Platform entrepreneurs	White-label knowledge chat for their own customers

ASK supports two levels of operator:

• Tenant admins — manage their org's documents, users, roles, and customer plans
• ASK platform admins — Logic Spine operators who approve signups, configure LLM providers, and manage subscription plans

Core capabilities

Document ingestion and knowledge base

Upload PDFs, DOCX, PPTX, TXT, and more. ASK:

• Extracts text and table-of-contents structure
• Chunks content in a structure-aware way (not naive fixed-size splits)
• Embeds chunks into pgvector for semantic search
• Tracks document status: uploaded → processing → verified

Documents must be verified before they enter the live knowledge base — so admins control what the AI is allowed to cite.

Hybrid RAG retrieval

When a user asks a question, ASK does not rely on vector search alone. Default retrieval combines:

1. Vector search — semantic similarity on embeddings
2. BM25 full-text search — keyword and phrase matching
3. TOC routing — uses document headings to pull the right sections
4. RRF fusion — merges ranked results into a single relevant context

This hybrid approach reduces missed answers when users phrase questions differently from how documents are written.

Streaming chat with citations

Answers stream in real time (SSE). Assistant messages include source citations — users can see which document chunks supported the response. That matters for trust, compliance, and debugging wrong answers.

Multi-tenant RBAC

Each organisation has its own users, roles, and permissions:

Role	Typical access
Super admin	Full tenant control
Admin	Users, docs, billing, settings
Knowledge manager	Upload, verify, prompts
Editor	Upload docs, chat
Viewer	Chat only

Permissions can be overridden per tenant. Features can also be gated by subscription plan modules (documents, users, permissions, etc.).

Public tenant chat

Share a branded chat link for end users who only need Q&A — not the admin dashboard. Ideal for customer support portals, member-only help, or lightweight product assistants.

Web and mobile

• Web app — Next.js dashboard for tenant admins and platform operators
• Mobile app — Flutter client for chat, admin, and ASK operator workflows on the go

Optional agentic / MCP layer

For advanced setups, ASK supports an MCP (Model Context Protocol) microservice for tool-based agent flows — e.g. structured knowledge-base search as a callable tool. Standard production chat today uses the streaming RAG endpoint.

How ASK works: end-to-end flow

1. Organisation signup

A user signs in with Google, requests a new organisation, and waits for platform approval (signup is approval-gated for quality control). Once approved, they become tenant admin with a fresh workspace.

2. Upload and verify documents

Admin uploads files → system extracts text → admin verifies → background worker chunks and embeds → knowledge base goes live.

3. User asks a question

Chat request → query embedding → hybrid retrieval → context assembly → LLM streams answer → message saved with citation sources → usage metered against plan budget or wallet.

4. Subscription and billing

ASK has two billing layers:

• Platform → tenant — organisations subscribe to ASK plans via Razorpay (INR); included LLM budget per cycle with wallet overage optional
• Tenant → end customer — tenants can sell their own plans to their users with separate Razorpay keys

This lets ASK work both as internal tooling and as infrastructure for your own AI product.

ASK vs building your own RAG stack

	DIY RAG	ASK
Time to first chat	Weeks–months	Days (after approval)
Multi-tenancy	You build it	Built-in
Hybrid retrieval	You tune it	Default pipeline
Auth & RBAC	You build it	Google OAuth + roles
Billing	You integrate	Razorpay subscriptions
Mobile	Extra project	Included

Logic Spine still builds custom RAG when your workflow is unique — ASK is for when the pattern is standard: documents in, grounded answers out.

ASK vs generic ChatGPT

Generic models are broad but ungrounded. They invent policies, quote wrong numbers, and cannot cite your PDFs.

ASK grounds every answer in retrieved chunks from your verified documents. Wrong answers are traceable to sources — and fixable by updating the knowledge base.

Technology under the hood (high level)

ASK is a production monorepo, not a prototype:

Layer	Stack
API	Python, FastAPI, SQLAlchemy async
Database	PostgreSQL 16 + pgvector
Queue	Redis, RQ for document workers
Frontend	Next.js 14, React, Tailwind
Mobile	Flutter
LLM	Configurable providers (OpenAI, Google, Anthropic, Groq, Mistral) via admin console
Storage	S3-compatible document storage
Payments	Razorpay
Deployment	GCP Cloud Run

LLM provider keys are managed in the platform admin console — not scattered in environment files — so operators can swap models without redeploying code.

Who operates the platform?

ASK Admin is the Logic Spine operator console: approve tenant signups, configure LLM providers and models, manage platform plans, and monitor billing. Tenant users never see this layer — they get their own /app dashboard.

ASK in the Logic Spine family

Logic Spine ships three flagship products:

Product	Domain	Solves
Sumikar	sumikar.logicspine.com	Carbon compliance — CBAM, BRSR, ICM
ASK	ask.logicspine.com	Document-grounded AI — RAG chat for teams
UnderKB	underkb.logicspine.com	Exact KB file compression for portals

Same philosophy across all three: find a specific, recurring pain point and ship focused software — not a bloated suite.

Frequently asked questions

Can I embed ASK chat on my website?

Public tenant chat at /t/{slug}/chat is the lightweight end-user experience. Deeper embed and API access are on the roadmap — contact us for custom integration needs.

What file types are supported?

PDF, DOCX, PPTX, TXT, and more via the document pipeline. Upload limits depend on plan tier.

Is my data shared across tenants?

No. ASK is multi-tenant with isolation — each organisation's documents and chats are scoped to that tenant.

Can I use ASK for customer support?

Yes. Many teams use ASK as an internal copilot first, then expose public chat to customers with RBAC and optional tenant-side subscription plans.

Getting started

1. Visit ask.logicspine.com
2. Sign in with Google and request an organisation
3. After approval, upload and verify your first documents
4. Open chat and ask a question only your docs can answer
5. Share /t/your-org/chat with users who need read-only access

Need a custom RAG workflow — ERP hooks, proprietary models, on-prem deployment? Contact Logic Spine. We built ASK because the pattern was universal; we build custom when your pattern is not.

ASK — Upload. Ask. Answer.