Company fact sheet

KOREV AI at a glance

Updated:

In short

KOREV AI is a French company based in Saint-Martin-d'Hères, Isère, that builds custom AI agents and the sovereign infrastructure that runs them: Atlas, a model built on Olmo 3 32B and adapted to each client's data, installed on-premises or hosted in France, and PRISM, which coordinates the agents and controls critical processing, subject to human approval.

This page gathers verifiable information about the company, its products and what it does not claim.

Identity

Full legal information: legal notice.
FieldValue
Company nameKOREV AI
Legal formFrench simplified joint-stock company (SAS)
SIREN107 888 539 (Grenoble Trade and Companies Register)
Registered office13 rue Marcel Chabloz, 38400 Saint-Martin-d'Hères, France
President and founderAmine Mohamed
Official websitehttps://www.korev-ai.com
Contactcontact@korev-ai.com
Website languagesFrench, English

Not to be confused with Kore.ai. KOREV AI is an independent French company, unrelated to Kore.ai, a US software vendor with a similar name.

What KOREV AI does

KOREV AI designs, installs and maintains AI agent systems for companies and organisations: business agents connected to existing tools (email, Odoo, business software) for accounting, sales, administration, logistics and marketing; custom applications and SaaS; and the infrastructure that runs them. Work cited: DICA France (decor visualiser and business agents) and Inovallée (tools for monitoring companies in its network, community management and marketing).

The principle: an AI answer without a source or a validation step cannot be used in a regulated process. KOREV systems keep the source of each piece of information, apply validation rules defined by the organisation and leave the final decision to an authorised person.

Products

ProductRole
KOREV AtlasEnterprise reasoning model built on Olmo 3 32B (an open model released by the Allen Institute for AI, Ai2), adapted to each client's data and installed on a dedicated machine (MSI EdgeXpert or AMD Ryzen AI Max) or hosted in France. The adapted model belongs to the client
KOREV EvidenceWorkspace where specialised agents turn a case file into a sourced deliverable: steps followed, source of each piece of information, report exportable for an audit or a committee
KOREV PRISMCentral AI that coordinates the agents and controls their work. For critical processing, several models compare their conclusions, rules and a quorum apply, and the result is held for human arbitration when the conditions are not met
KOREV PulseConsultant-led assessment to choose a first process and define its control requirements
AI Act assessmentSelf-assessment of exposure and action plan; guidance only, with no legal analysis or certification
KOREV LabsResearch on embodied AI, robotics and physical intelligence. A neurosensory tissue for training AI models, covered by a filed patent, developed in partnership with CNRS, Inria, Université Grenoble Alpes and CEA

How the three products fit together is described on the Platform page.

Deployment options

  • On-premises: Atlas and the agents run on a dedicated machine at the client's site (MSI EdgeXpert or AMD Ryzen AI Max machine), supplied by KOREV or bought by the client, installed and maintained by KOREV.
  • Cloud in France: the same system hosted in France, for example with OVHcloud for DICA France.
  • Atlas is adapted to the client's data on ISO 27001-certified hosting infrastructure, with transfers over secure channels. The adapted model belongs to the client.
  • External models are used only if the client chooses them; they are then listed explicitly.
  • KOREV does not use its clients' data to train public models.

Source: Trust Center.

Published research

KOREV AI has published a technical note on verifiable decision records (version 1.0.2). It describes a format that lets a reviewer detect that a decision history has been altered, truncated or left incomplete, and that classifies each record as VERIFIED, INCOMPLETE or FAILED. The note states what the format does not prove: regulatory compliance, the truth of the content, or that a human review actually took place.

Read the note: Verifiable decision records.

What KOREV AI does not claim

  • No security certification (ISO 27001, HDS, SecNumCloud or other) is claimed by KOREV AI to date. The ISO 27001 certification mentioned for adapting Atlas belongs to the hosting infrastructure used.
  • No guarantee of AI Act compliance. KOREV systems are designed for regulated environments and make some requirements easier to implement (traceability, human oversight, documentation). Compliance depends on the system, its use and the organisation.
  • No promise of error-free output. Comparing several models reduces dependence on a single model; it does not remove errors.
  • No KOREV Atlas performance benchmark is published until it has been rerun on a stable public protocol.

Frequently asked questions about KOREV AI