# ClosyTalk

> ClosyTalk is an enterprise AI platform by Beliven Srl (Udine, Italy) that gives companies private, GDPR and EU AI Act compliant access to multiple Large Language Models through a single interface. It is not a SaaS: every customer gets a dedicated instance installed inside their own perimeter.

## Distribution model

ClosyTalk is not a multi-tenant SaaS. Customers do not buy access to a shared platform. Each deployment is a dedicated single-tenant instance:

- **On-premises (native model)**: software, models and data on the customer's own hardware, inside their network, up to full air-gap operation with no internet access.
- **Customer cloud**: the same dedicated instance running on the customer's Azure, AWS or Google Cloud tenant.
- **ClosyTalk cloud**: an isolated environment in European data centers, with maintenance, updates and scaling handled by ClosyTalk.

What never changes: the environment is always one customer only.

## Who it is for

ClosyTalk is designed for organisations from roughly 20 office-based employees upward that meet three conditions:

1. Enough people working on documents that knowledge starts to disperse across departments, folders and mailboxes.
2. Documentation that cannot leave the company, such as contracts, client data, projects and case files.
3. Knowledge that already exists internally but is not searchable without knowing which file to open.

The selection criterion is not the industry but the density of documentary knowledge.

## Three knowledge profiles

- **Regulatory and contractual knowledge**: long texts, multiple versions, differences that matter. Typical assistants compare contract versions with a table of differences, query practice notes and regulatory collections in PDF, and flag risky clauses with the exact reference.
- **Technical and operational knowledge**: manuals, specifications, procedures, spreadsheets. Typical assistants extract insight from Excel files with correctness checks, make technical documentation searchable with page-level references, and expose internal procedures to every department.
- **Commercial and relational knowledge**: how the company writes and decides. Typical assistants draft communications aligned to the company tone of voice, produce executive summaries with action points from meeting minutes, and structure business cases with NPV, IRR and payback.

## Sectors

Dedicated pages exist for manufacturing and service SMEs, law and tax firms, and accounting firms. The same architecture is applied to service companies, energy and multiutility, labour consultants, engineering and design, construction and public tenders, insurance and brokerage, consulting firms, real estate and property management, logistics and transport.

## Key features

- **Multi-model access**: GPT-5.6, Claude Opus 5, Qwen 3.7, Mistral and other open-weight models from one interface. New models enter the platform with zero migration and zero downtime, and the interface stays identical.
- **Private knowledge base**: connects SharePoint, Google Drive and company repositories. Each assistant can be bound to its own knowledge base, and every answer carries granular references to the source document and page.
- **Custom agents**: assistants configured on the company's own processes, with a defined role, domain, objectives and behavioural limits.
- **Enterprise governance**: role-based granular permissions, complete audit trail of every interaction, SSO and LDAP integration, per-team AI budgets with automatic alerts.
- **Responsible AI**: human confirmation required before exporting critical output, reading time estimates to discourage blind acceptance, and admin-defined language and compliance policies applied to every interaction.

## Compliance

- Compliant by design with GDPR and the EU AI Act.
- No company content is used to train third-party models.
- End-to-end encryption and verifiable data sovereignty.
- Complete audit trail, ready for internal audits and regulatory inspection.
- Relevant to organisations subject to NIS2, to Italian Legislative Decree 231/2001 organisational models, and to professional secrecy obligations.

## Project method

Projects follow four phases and take roughly four months from kickoff to assistants in production: analysis of needs with each business function, secure setup on the customer's own policies, immersive in-person training in groups, and co-design of the agents together with the teams.

## News

The site publishes analysis on AI regulation and enterprise adoption at https://www.closytalk.com/news, covering the EU AI Act and its Italian implementing decrees, governance and privacy, cybersecurity, models and research, data centre infrastructure, and business adoption.

## Company

Beliven Srl
Viale Del Ledra, 56, 33100 Udine (UD), Italy
VAT: IT02950460309
Phone: +39 0432 169 8253
Email: info@closytalk.com
Website: https://www.closytalk.com