Home / Tools / KEEL

The trusted context layer for enterprise AI.

KEEL turns the documents and data scattered across your organization into evidence-backed answers, retrieved through chat and a REST API, with a confidence score and sources on every response. It is a ViitorCloud infrastructure accelerator we deploy inside your engagement.

See how it works
Files up to 500 MB · 1,000-page documents · Tens of thousands of docs
A context layer, not another chatbot

Chat is one interface on top of KEEL. Underneath is a context layer, ingestion, tagging, embeddings, a knowledge graph, and a REST API, that any tool can query. The chat UI and the API run the same retrieval engine, so every copilot, assistant, and internal tool is grounded in the same evidence-backed context. Deployed inside ViitorCloud engagements as an infrastructure accelerator.

Evidence, not vibes

Every answer carries its receipts.

Ask a question in plain language. KEEL streams the answer, then shows the exact source chunks behind it, each with a match score, and a single confidence figure from the retrieved evidence. Nothing to take on faith.

app.keel.io
What's our refund policy for enterprise plans?

Enterprise plans are eligible for a prorated refund within 30 days of renewal. The amount is calculated from the cancellation date forward, per the master agreement.

High confidence — 87%Evidence · 3 sources
FY24 Enterprise Agreement.pdf92%
§ 7.2 — Refunds

Enterprise customers may receive a prorated refund within 30 days of renewal…

Support Playbook.docx81%
Billing › Cancellations

For annual plans, refunds are calculated from the cancellation date forward…

Master Services Agreement.pdf74%
Clause 12 — Termination

Either party may terminate for convenience with 30 days written notice…

The problem

Enterprise AI is only as reliable as the context behind it.

The gap isn't a lack of AI. It's a lack of trustworthy context to give it.

Knowledge is fragmented

Documents, drives, tickets, and reports pile up across dozens of tools with no shared structure.

Verification takes longer than the work

People re-check facts by hand because they can't trust what a search result or AI answer tells them.

AI retrieves without proving

Most retrieval-augmented tools return a plausible chunk of text, not a traceable source.

Institutional knowledge walks out the door

When people leave, the context they carried in their heads leaves with them.

Generic RAG breaks down under scrutiny

Simple chunk retrieval is not enough when an answer may need to be defended or reviewed.

Every AI project starts from zero

Without a reusable context layer, each new copilot or assistant rebuilds its own retrieval pipeline.

The platform

One backbone for your organization's knowledge.

KEEL takes data from anywhere in your organization and makes it usable by AI, structured, searchable, and backed by evidence.

Connect

Connect your data

Upload PDFs, Word docs, spreadsheets and text, or connect Google Drive and hand-pick exactly which files enter KEEL. Files scale to 500 MB.

Process

Process at scale

An asynchronous pipeline parses with Docling, generates AI tags, chunks the text, embeds it into a vector store, and maps entities into a graph.

Retrieve

Retrieve with evidence

Ask in chat or call the REST API. Every answer lists the source chunks it drew from and carries a confidence score from the retrieved matches.

How it works

From raw file to retrievable knowledge.

Upload or sync a file and KEEL takes over: an asynchronous, resumable pipeline parses, tags, chunks, embeds and graphs it, then it's ready to answer questions.

01
Upload
PDF · DOCX · XLSX
02
Parse
Docling
03
Chunk
512 tokens
04
Embed
Qdrant
05
Graph
Neo4j
06
Retrieve
Chat · API
Asynchronous and resumable, large files stream page by page, so a stalled upload never restarts from zero.
Knowledge graph

Every document, mapped into a graph.

KEEL runs LLM-based entity extraction over each document and upserts the results into a Neo4j graph, people, organizations, projects, products, industries and dates, deduplicated across your workspace and linked back to the chunks that mention them.

When your question names a known entity, graph-augmented retrieval pulls in extra context from the documents connected to it.

MENTIONSAUTHORED_BYBELONGS_TOREFERENCES
Document
Person
Organization
Project
Product
Industry
Date
Capabilities

Everything you need to make data answerable.

A focused enterprise feature set, from ingestion to retrieval to programmatic access.

Evidence-backed chat

Answers stream in token-by-token, then surface a confidence indicator and an evidence panel of the exact source chunks behind the response.

AI document tagging

Each document is tagged automatically by an LLM at ingestion. Tags stay fully editable and power search and filtering across your data.

Vector + graph storage

Chunks are embedded into a Qdrant HNSW index for fast similarity search, while entities and relationships are mapped into a Neo4j graph.

REST API

Search, context, chat and ingestion endpoints let your own tools push data in and pull evidence-backed context out, the same engine as the UI.

Source connectors

Connect Google Drive via OAuth, browse the folder tree, and select the specific files to ingest. OneDrive support is on the way.

API keys & rate limits

Generate workspace-scoped read-only or read-write keys, cap them at 100 requests per minute, and revoke any key the moment you need to.

Workspace dashboard

Live counters, a recent ingestion activity feed, and a pipeline health view tell you what data is available and how the platform is being used.

Roles & permissions

Admin and Standard roles govern administrative actions, while every member of a workspace can retrieve from the full ingested dataset.

AI, end to end

Real models doing real retrieval work.

KEEL is AI-first by design, every step from parsing to answering runs on production models, not heuristics.

Docling parsing

Extracts clean text, headings and tables from PDF, DOCX, TXT, CSV, XLSX and PPTX, streaming large files page by page.

LLM document tagging

Generates descriptive tags from each document at ingestion to organize and filter your corpus.

OpenAI embeddings

text-embedding-3-small produces 1,536-dimension vectors, indexed with HNSW and cosine similarity for fast search.

Entity & relationship graph

LLM-based extraction maps people, organizations, projects and more into a Neo4j graph, deduplicated per workspace.

Graph-augmented retrieval

When your question names a known entity, KEEL pulls in extra chunks from documents linked to it.

Confidence scoring

Every answer reports the mean similarity of its top retrieved chunks as a percentage, so you know how grounded it is.

Architecture

Engineered for real enterprise data.

KEEL is engineered for large, messy, real-world enterprise data, not a demo dataset.

Parsing & ingestion

Docling parses PDF, DOCX, TXT, CSV, XLSX and PPTX through an asynchronous, resumable pipeline, streaming large files page by page.

Storage

PostgreSQL holds relational data, Qdrant indexes embeddings with HNSW and cosine similarity, and Neo4j maps entities and relationships.

Models

OpenAI's text-embedding-3-small (1,536 dimensions) for embeddings and gpt-4o-mini / gpt-4o for chat, fixed per workspace so retrieval stays consistent.

Integration

A REST API mirrors every UI capability, search, context assembly, chat, and ingestion, so external tools use the same engine as KEEL itself.

Connectors

Google Drive today via OAuth and manual file selection, with more sources planned.

Scale

Built for files up to 500 MB, thousand-page documents, and workspaces with large, growing document counts.

500 MB
Max file size
1,000+
Pages per document
1,536
Embedding dimensions
100/min
API requests per key
Use cases

Built for how enterprise knowledge actually gets used.

The same context layer supports every one of these, no separate tool per use case.

Enterprise knowledge search

Problem

Employees re-search for information that already exists somewhere in the org.

KEEL

One searchable layer across every uploaded and connected document.

Outcome

Faster answers, less duplicated research.

Internal AI copilots

Problem

Copilots give generic answers when they lack real organizational context.

KEEL

A REST API that grounds any internal assistant in KEEL's evidence-backed context.

Outcome

More useful internal tools without building retrieval from scratch.

Policy and SOP intelligence

Problem

Long policy documents are slow to search and easy to misquote.

KEEL

Plain-language questions answered from the exact policy text and section.

Outcome

Fewer errors from outdated or misremembered procedures.

Project and delivery knowledge

Problem

Project history and decisions scatter across files and get lost when teams change.

KEEL

A structured, searchable record of project documents anyone can query.

Outcome

Less institutional knowledge lost to turnover.

Customer support knowledge

Problem

Support teams re-research issues because the knowledge base is hard to search.

KEEL

Unified retrieval across product docs and support material, with source snippets.

Outcome

Faster resolution, more consistent answers.

Technical documentation

Problem

Engineers spend real time hunting through scattered, versioned docs.

KEEL

Semantic search that understands meaning, not just keyword matches.

Outcome

Less time searching, more time building.

Regulated industry knowledge

Problem

Teams need to justify where an answer came from; generic AI can't show its work.

KEEL

Every answer ships with a confidence score, source documents, and snippets.

Outcome

Answers that can be reviewed, not just trusted blindly.

AI application grounding

Problem

Teams building AI apps need reliable context without a custom retrieval stack.

KEEL

A REST API for search, context assembly, and chat, the same engine as the UI.

Outcome

Faster time-to-build for AI-powered applications.

Institutional knowledge preservation

Problem

Undocumented knowledge leaves when experienced employees do.

KEEL

Continuously ingested, structured knowledge that persists independent of any one person.

Outcome

Reduced knowledge loss, faster onboarding.

Who it's for

Built for different roles, one shared context layer.

Give every AI initiative a shared foundation instead of one-off pilots.

One workspace-scoped context layer that any internal tool can query through the REST API.
Centralized control over users, roles, and API key access.
Evidence attached to every answer, so AI-assisted decisions can be reviewed, not just trusted.
Why KEEL

Answers your team can actually trust.

KEEL is built to be evidence-first, secure, and ready for the scale of real enterprise data.

Evidence on every answer

No black-box responses. Each answer shows the source chunks it used and a confidence score derived from retrieval similarity.

Built for enterprise scale

Designed from day one for files up to 500 MB, thousand-page documents and hundreds of thousands of vectors per workspace.

Workspace-scoped by default

Every search is filtered to your workspace. Access tokens live in memory, refresh tokens stay in HttpOnly cookies.

Programmatic and UI parity

The REST API runs the same retrieval engine as the chat UI, so external tools get identical, evidence-backed context.

Differentiation

More than document search. More than basic RAG.

KEEL isn't competing with a single chatbot or search box, it's built to be the context infrastructure underneath all of them.

CapabilityKeyword searchBasic RAGCopilot add-onsKEEL
Entity & relationship (graph) awareness✓ Included
Evidence & source attribution per answer✓ Included
Reusable across tools (REST API-first)✓ Included
AI-generated tags & metadata✓ Included
Fact-level validation & contradiction detection◌ Roadmap
Security & access

Built with security and traceability in mind.

Workspace isolation

Every search, document, and API key is scoped to a single workspace, there is no cross-workspace data access.

In-memory access tokens

Access tokens live in memory only, never in localStorage. The refresh token stays in an HttpOnly cookie the browser controls.

Two-factor authentication

Add an authenticator-app second factor to any account, with one-time recovery codes if you lose access to it.

Scoped, revocable API keys

Generate read-only or read-write keys, cap them at a configurable rate limit, and revoke any key immediately.

Source traceability

Every answer names the documents and snippets it came from, so it can be checked against the original content.

Admin and Standard roles

Administrative actions, managing users, connectors, and API keys, are separated from day-to-day retrieval.

Roadmap capabilityComing next

From context retrieval to trusted knowledge.

KEEL won't treat every document statement equally. It will validate information before promoting it into trusted organizational knowledge.

01
Relevance check
02
Fact extraction
03
Entity canonicalization
04
Dedup & corroboration
05
Conflict detection
06
Confidence routing
Confidence routing sends each fact to one of
Trusted store
Promoted to trusted knowledge
Human review
Contested facts routed to a person
Dropped
Irrelevant or unverifiable, discarded
Positioning

Built for teams that can't afford to guess.

Architecture-first, not a retrofitted chatbot demo
Built for enterprise deployment, not single-user tools
API-first, every UI capability is also a REST endpoint
Built with high-trust environments in mind from day one
Designed to feed copilots and assistants, not replace them
FAQ

Questions we hear most

Anything else, email the practice.

No. Chat is one interface on top of KEEL. Underneath is a context layer, ingestion, tagging, embeddings, a knowledge graph, and a REST API, that any tool can query. The chat UI and the API run the same retrieval engine.

Basic RAG retrieves similar chunks and hopes for the best. KEEL adds AI-generated tags, an entity and relationship graph for entity-aware retrieval, source attribution on every answer, and a confidence score derived from retrieval similarity.

Today you can upload files directly (up to 500 MB) or connect Google Drive via OAuth and hand-pick which files to ingest. OneDrive and additional connectors are planned.

Yes. Each answer lists the exact source chunks it drew from, document name, section, and snippet, plus a confidence percentage, so it can be checked against the original content.

Yes. The REST API mirrors every UI capability, search, context assembly, chat, and ingestion, so copilots, assistants, and internal tools can be grounded in the same evidence-backed context.

A roadmap capability that validates information before promoting it to trusted knowledge: relevance checks, fact extraction, entity canonicalization, dedup and corroboration, conflict detection, and confidence-based routing to a trusted store, human review, or rejection.

Yes. LLM-based entity extraction runs at ingestion and populates a Neo4j graph. When a question names a known entity, retrieval pulls in extra context from the documents connected to it.

CIOs and CTOs standardizing AI initiatives, AI and data teams building on a ready retrieval stack, business teams who need answers with receipts, high-trust sectors like defence and government, and partners shipping AI solutions for clients.

Make your organization's knowledge answerable.

Create a workspace, connect your data, and start retrieving evidence-backed answers, deployed inside your engagement.

Explore Viitor Atlas
Evidence on every answer · Chat + REST API · Workspace-isolated · Yours to keep