Not a chatbot. An assistant that cites its sources.
Sofia chats to your data in plain English, drafts disclosures, sketches tickets, suggests setpoints — and shows you where every number came from. Human approval at every step.
| Sede Lisboa | −24,1% | gas boilers → heat pump |
| Centro Porto | −19,8% | EV fleet +12 units |
| Logística Setúbal | −12,3% | shift to off-peak hours |
Eight specialists, six with a persona
Eight specialists, six with a persona
Sofia (sustainability), Luna (energy), Atlas (BMS), Vega (value chain), Lumen (retail), Aura (compliance) — plus two more dedicated to product surfaces. Same chat, different lenses.
Sources in every reply
Each answer links straight to the source table, metric or document. Nothing comes out of thin air — and the auditor can follow the trail.
Writes to your data
Eleven writable tables. Inserts manual metrics, updates disclosures, flags incidents — always with human approval.
Persistent memory
Remembers last quarter’s thread when you ask something this quarter. No re-briefing.
Guardrails per role
Pick what each persona can do alone, what needs sign-off, what’s off-limits. Configurable policy.
Immutable audit log
Every interaction: input, context, output, model, cost. 12-month retention, exportable for audit.
In 3 steps
Plug in the data
BACnet, ERP, Excel, APIs — Sofia indexes the lot inside hours.
Set the tone
Pick the personas you need, tune guardrails, scope by role.
Put her to work
Within two weeks the team is asking her things in English, Portuguese or Spanish — answers come with sources.
Sofia gives the team back about 14 hours a week, just on compiling reports. It was the most tedious job nobody wanted.