Hades

An autonomous supplier due-diligence agent. Six research nodes check sanctions (OFAC, UN), the company registry, LkSG and CSDDD signals, ESG and labour risk, news sentiment and Hermes market intelligence. Claude scores each dimension from 1 to 10; tested deterministic code turns those scores into the verdict.

The live agent is paused since November 2026. Below: a real investigation from May 2026, and the decision logic itself, running in your browser.

Hades investigation of Bosch: risk score 5.0, risk level High, the six research steps completed — A real investigation, run on 16 May 2026 through the live agent. Back then the model also set the risk band and called a 5.0 "High". Since 4 July 2026 the code below decides the band, and 5.0 is Medium.

The model advises, the code decides

Set the six dimension scores the model would return and the hard facts the research nodes found. The weighted score, the risk band, the recommendation and every override come from a direct port of agent/scoring.py, checked against the Python original on 20,000 random inputs. The example inputs are illustrations, not real companies.

Confirmed priority hit on a sanctions list Registry shows the company dissolved or insolvent LkSG/CSDDD research returned a red flag A sanctions list could not be fetched (screen unverified)

decided by: deterministic code, never the model

Code, tests and the research nodes: github.com/eugnmueller-87/hades