Spektr

Nine named agents research a company, verify that it operates, classify its industry and assess source of funds, and an agent builder lets customers define more. Without the models, only the configurable process layer remains, the very state of the market Spektr was founded to move past.

A builder prompt sets the goal, research gathers information, customer defined routing branches the workflow, and at designated points the agent pauses for a user to choose among up to five configured actions, recorded as the human decision. Agents can be shipped, watched in production and rolled back immediately, though the pause points are set by the customer rather than a published default.

Agent performance is tracked by a monitoring layer and research results are published to the customer profile, but no accuracy figure, false positive rate or validation pack is published.

Case studies name Nexi, Pleo and Santander Leasing, and the head of compliance at Fyorin is quoted by name and title, but no measured outcome is attached to any of them. Funding round coverage in 2026 offers only generic claims about hours becoming minutes.

Customers are promised full control over their data and how AI operates on it, yet whether documents, analyst decisions or agent outputs train models served to others is not stated. A shared agent library and an optimization loop that refines agent logic make that question live.

Privacy certification under ISO/IEC 27701:2019 followed a Stage 2 audit by Mastermind Assurance in June 2026, and the trust page states GDPR Article 28 compliance with data processing agreements across subprocessors. The privacy policy covers website visitors and business contacts only, and no subprocessor list or retention schedule appears.

ISO/IEC 27001:2022 and a SOC 2 Type II attestation are named on the trust page, under an integrated management system with annual re audit. Certificate numbers, the SOC 2 auditor and downloadable reports do not appear there.

Voluntary international standards audited by private bodies make up its credentials, which is assurance rather than regulatory standing. As a software supplier it states no license, registration or supervised sandbox.

Its AI management system holds ISO/IEC 42001:2023 certification, from data governance to accountability. Classifying industries and judging source of funds will read some sectors and countries less reliably than others, yet no evaluation across jurisdictions or business structures is published.

Recorded human decisions at each pause point give an institution a trail, but Spektr publishes no commitment behind an agent's output. Wrongly classified companies, or those judged inactive, may be declined without learning an agent produced the finding.

Agents run on unnamed models: neither the help center nor the trust page names a provider, family or version. Nor is it stated which providers receive customer documents and corporate records or what they retain.

Developer documentation for a public API, an integrations page, release notes and event orchestration that lets outcomes in one system trigger checks in another are all published. Which customer systems connect natively, and which data providers sit behind the agents, is not identified.

Data is handled and hosted within the European Union, per the AWS Marketplace listing and the homepage. Cloud provider, region and any single tenant option are not named, nor where model inference runs.

An AWS Marketplace listing from spektr ApS publishes three 36 month tiers: Professional at $750,000, Enterprise at $4,500,000 and Premier at $10,500,000. What volume or agents each tier includes is not stated, and the company's own site carries no pricing.

Buyer types range from a global bank's leasing arm and a Danish small business bank to payments infrastructure and an electric vehicle charging platform buying onboarding compliance. Material is organized by product rather than institution type, so fit is judged from customer stories.