What is TensorLot?

What problem does TensorLot solve?

Most enterprises sit on millions of documents: five versions of the same procedure, copies in forgotten sites, files from 2019 that still rank first in search. When an AI assistant or an AI agent is pointed at that raw document landscape, it answers with confidence — from the wrong file. Gartner names poor data quality the #1 obstacle in generative-AI projects (2025).

TensorLot addresses the data underneath the AI, not the model on top of it: it finds what is contradictory, duplicated or outdated, helps teams fix it, and makes sure that assistants and agents read the canonical, current version — with a citation.

How does TensorLot work?

TensorLot Core runs in four stages. Nothing moves, nothing is rewritten behind your back.

1. Connect. Sources are connected read-only: SharePoint sites, libraries and site collections, Confluence, Jira, GitHub. Sync is incremental and permissions are mirrored from the source system. There is no migration and no copy to maintain.
2. Analyse. Every document receives a health score (0–100), and so does the whole tenant. TensorLot detects contradictions across documents and sources, outdated content and duplicates — each recorded as a finding with severity, status and the affected documents, and a canonical candidate where one exists.
3. Clean. Findings come with a recommended action (resolve, archive, move, ignore). Action rules define once what “archive” or “resolve” means per URL prefix, library, site or tenant — the most specific rule wins. TensorLot never writes to SharePoint itself; the person or agent acts with their own rights.
4. Serve. The cleaned knowledge is served to people through Tensor Bot — grounded answers with citations, scoped to a knowledge space — and to AI agents through one MCP endpoint with 11 governed tools. Every call is audit-logged with agent, tenant and tool.

What does “Knowledge Tensor” mean?

The Knowledge Tensor is TensorLot’s name for the curated knowledge layer that TensorLot Core builds and keeps in sync on top of the raw sources: one representation of the document landscape across several dimensions at once — documents and their versions, where they live, how healthy they are, how they relate to each other (duplicates, contradictions, canonical versions), and who may see them.

Everything TensorLot serves reads from this layer instead of from the raw file share: Tensor Bot answers from it, the MCP server exposes it to agents, and the console shows it as health scores, findings and contradictions. That is why answers can cite the canonical version and why agents get one clean context instead of five conflicting files — which is also where the measured reduction in tokens per answer comes from.

Who is TensorLot for?

TensorLot is built for enterprises and public-sector organisations whose knowledge lives in Microsoft 365 / SharePoint, Confluence, Jira and similar systems, and who want to run assistants and AI agents on that knowledge without inheriting its inconsistencies.

• Knowledge-management, information-governance and compliance teams that need to know what is outdated, duplicated or contradictory — and fix it with rules rather than by hand.
• AI, data and platform teams that build assistants or agents (Copilot Studio, Claude, OpenAI, LangGraph, custom frameworks) and want one governed, cited retrieval layer instead of raw document dumps.
• Business functions such as procurement, compliance, onboarding and support, where people ask the same questions of the same documents every day.
• Regulated organisations that require read-only access, EU hosting, a clear governance model for agents and a full audit trail.

TensorLot and AI agents (MCP)

TensorLot is an MCP server. MCP (Model Context Protocol) is the open standard that lets an AI agent call external tools. An agent is created in the TensorLot console, scoped to a defined knowledge space (sites, libraries or topics), and given a default-deny tool list — tools are off until switched on. It connects through one endpoint with one key, which expires and can be revoked at any time.

The MCP server exposes 11 governed tools, among them ask and search_documents (semantic search and grounded, cited answers — the same retrieval as Tensor Bot), list_findings and get_finding (quality findings such as duplicates, contradictions and outdated content), and set_document_action (governed actions such as archive or resolve, executed with the user’s rights and audit-logged). A playground lets teams dry-run a task over the real MCP endpoint and inspect every tool call before going live.

How does TensorLot compare to Microsoft 365 Copilot, Glean and RAG frameworks?

TensorLot is not an assistant that competes with Copilot or Glean, and not a vector database or RAG framework. It is the layer underneath them that makes the data itself trustworthy — and it can feed any of them.

In short: Copilot and Glean answer over your knowledge as it is; TensorLot makes sure that what they (and any other model or agent) read is deduplicated, contradiction-checked, current — and cited.

Security, hosting and compliance

• Read-only, permission-aware access to sources; TensorLot never writes to SharePoint or any other source system.
• Hosting in the EU; private cloud or deployment in the customer’s own VPC on request. Encryption in transit and at rest.
• Agents are part of the governance model: scope, accountable owner, expiring keys, default-deny tools and a full audit log per call.
• Compliance status (August 2026): GDPR by design (EU hosting, DPA, deletion on request); SOC 2 and ISO 27001 in preparation; security questionnaires and architecture walkthroughs on request.

How is TensorLot priced?

There are three ways in: a Proof of Concept (typically 2–4 weeks on the customer’s own documents, with a full quality report), Knowledge Management (the full knowledge layer in production: Tensor Bot with cited answers, governed clean-up, first agents via MCP) and Enterprise (organisation-wide deployment, dedicated support, private-cloud options). Pricing is based on the number of documents and data sources, users and the capabilities needed, and is quoted on request; POC costs are typically credited towards the first year.

Who is behind TensorLot?

TensorLot is the work of Dr. Omar El Sayed (Founder & CEO), following foundational research on enterprise knowledge management and AI-ready data quality, and Alexander Lichtenstein (CTO), who joined in November 2025. Scientific advisor is Prof. Dr. Dr. h.c. mult. Horst Stöcker; business advisor is Beschir Hussain. TensorLot GmbH is founded in Zug, Switzerland, in January 2027 — a Swiss company serving enterprises across Europe.

Product milestones: first prototype of TensorLot Core in January 2026 (ingestion, contradiction detection and quality scoring validated on real enterprise data); pilot-ready platform in April 2026; MCP server with governed tools, playground and action rules in August 2026.