ARIA

ARIA captures the floor operator's knowledge and puts it to work. Drop in the manufacturer's PDF, answer some calibration questions, and you're live in ten minutes. After that, ARIA continuously ingests everything that already exists — live signal trends, operator logbook entries, shift notes, machine failure history, and computed KPIs (OEE, MTBF, MTTR) — building a knowledge base that grows with every incident. When something goes wrong, five agents pass the problem like a real maintenance team passes a ticket: detection → diagnosis → work order → memory. Five Agents : - KB Builder — reads the manufacturer PDF with Opus 4.7 vision and captures the operator's floor knowledge through a short calibration dialogue. Live before they leave the terminal. - Sentinel — watches live signals against the KB; detects breaches, forecasts signal tails, and judges whether a drift warning is worth surfacing. - Investigator — the centrepiece. Diagnoses anomalies with Opus 4.7 extended thinking; writes and executes Python in Anthropic's sandboxed container to compute exact degradation rates from raw signal data; recalls and builds on every past failure. - Work Order Generator — turns the RCA into a printed sheet: root cause, remediation steps, exact part number, intervention window. - Q&A — natural-language operator chat; hands off to the Investigator when a deep diagnosis is needed. All five agents share 17 MCP tools as their only path to the database. Generative-UI artifacts — charts, diagnostic cards, work orders — stream into the operator's chat as the agents work.