“Where are we with this supplier onboarding?”
Step 4 of 6, awaiting approval since Tuesday. Compliance screening passed; the bank details check is blocked on a missing certificate.
Answer drawn fromProcurementComplianceFinance
Conversational layer · Part of BRAIAN
Carol
Carol
Carol is the conversational layer of BRAIAN. Employees use voice or text to navigate the processes they work on — across business units, within their own permissions. Carol adapts to each person: their role, their systems, their vocabulary. Every question and every action is logged under the same policies that govern production workflows.
What can you ask?
Carol answers from the systems a process already runs on, and names where each part of the answer comes from.
“Why was this invoice escalated?”
FinAgent matched the invoice to a purchase order with a CHF 12,400 variance, above the CHF 10,000 approval threshold in policy v4.2, so it routed to the cost centre owner.
Answer drawn fromFinanceAudit log
“What does this contract clause commit us to?”
Clause 8.3 sets a 30-day cure period and auto-renewal unless cancelled 90 days before term. LegalAgent flagged it against the standard playbook.
Answer drawn fromLegalProcurement
How does it work?
Organisational knowledge layer
Carol reads the same process definitions, policies, connectors and audit records the runtime uses. There is no separate knowledge base to maintain and no copy of your operational data to keep in sync.
Permission-aware retrieval
Every retrieval runs under the asking person's identity. Records outside their permissions are not summarised, not cited and not counted — the answer is built only from what that person could already open themselves.
From conversation to governed process
When a question turns into an action — approve, escalate, re-run a step — Carol hands it to the runtime. The action executes under its versioned policy and lands in the audit log like any other run.
Voice and text
The same layer, the same permissions, two ways in.
- Text
- In the BRAIAN console, and in Microsoft Teams and Slack through the existing connectors.
- Voice
- Speech-to-text in the console and on mobile, for hands-free status checks and approvals within threshold.
- Languages
- English, Italian and German, including mixed-language questions about the same process.
- Context
- A conversation keeps the process it is about, so follow-up questions do not need to repeat identifiers.
Personalisation and boundaries
Carol personalises to each employee; it does not model them. The distinction is a design constraint, not a setting.
What Carol adapts to
- The role and permissions on the account asking the question.
- The systems and processes that role actually works on.
- The vocabulary of the business unit — its terms for documents, steps and thresholds.
- Preferred language, level of detail and output format.
What Carol does not do
- It does not build a profile of an employee's behaviour, productivity or performance.
- It does not answer on behalf of a person or act with anyone else's identity.
- It does not surface records outside the asking person's permissions, including in summaries.
- It does not execute an action that exceeds the policy in force for that process.
- Conversations are not used to train models; they are logged as records, retained under the same policy as process traces.
The rest of the runtime
See Carol answer on your own process
We connect one process in a working session and ask it the questions your teams ask today.