Maria Meet Agent
Living intelligence from every meeting. Real-time transcription, decision extraction, and evidence-linked minutes β from silent observer to voice copilot.
Meetings become structured intelligence
Not just recording. Extracting decisions, actions, and evidence from live conversations β with every claim traceable to its source utterance.
Real-time Transcription
Speaker-diarized, timestamped transcript with partial/final segments. Confidence scores and evidence references on every utterance.
Decision Extraction
Automatic detection of decisions, open questions, and topic shifts. Every extracted item links back to the exact spoken segment.
Action Tracking
TODO items with owner assignment, due dates, and status β all evidence-linked. No decision without a traceable origin.
Evidence Chain
SHA-256 hashed audio segments, chained audit trail, and tamper detection. Every minute item cites its source utterance.
Structured Minutes
Auto-generated chapters, topic summaries, and final artifacts in markdown and JSON. Draft β Live β Final lifecycle.
Living Sales Intelligence
Meeting insights flow directly into your agent pipeline. Customer objections, pricing signals, and commitment language β captured and structured.
Two phases. One philosophy.
Start silent. Earn trust. Then speak β only when gated, evidence-backed, and invited.
Silent Minutes
Phase 1 β VOICE OFF
Join any meeting as a silent observer. Listen, transcribe, extract β never interrupt. All output is consent-gated.
Voice Copilot
Phase 2 β VOICE ON
Speak only when called. Summarize decisions, clarify open questions, structure options β all gate-controlled and rate-limited.
From raw audio to governed artifact
10 specialized agents. Each with a single responsibility. Connected by evidence references and gate checks.
Meet Adapter
Join meeting, receive audio stream, emit participant events
Streaming Ingest
VAD, chunk alignment, time-stamped audio buffer
ASR Service
Partial β Final transcription with timecodes and confidence
Diarization
Speaker labels, participant ID mapping, manual correction
Minutes Engine
Topic chapters, decisions, actions, open questions
Evidence Layer
Span management β every claim links to source segment
Gate Engine
Consent, Scope, Export, Speak β all fail-closed
Storage
Encrypted audio fragments, sessions, permissions
Live UI
Real-time minutes viewer with evidence links
Audit Trail
Chained hashes, prompt records, tamper detection
See intelligence as it forms
Real-time transcript with auto-extracted decisions, actions, and open questions β all evidence-linked.
I'd like to raise the Q3 sales target by 15%
We'll need additional marketing budget β is that approved?
Budget approved at director level. Cap at Β₯3M
Is the product launch date confirmed for next Friday?
Confirmed. Suzuki, get PR materials ready by Thursday
Raise Q3 sales target by 15%
ref: 14:03:22Additional marketing budget capped at Β₯3M (director-approved)
ref: 14:04:01Prepare PR materials by Thursday
Final confirmation of product launch date
ref: 14:04:18Every action is gated
Four gates protect every meeting interaction. When any gate fails, the system stops β never proceeds without authorization.
Gate-Consent
Phase 1+Host or designated role confirms recording consent
Transcription and storage proceed normally
Generation continues but nothing is persisted β local temp only
Gate-Scope
Phase 1+Meeting title, participant domains, and sensitivity tags evaluated
Full transcript and minutes stored per policy
Summary only β verbatim transcription prohibited (e.g., external participants)
Gate-Export
Phase 1+Explicit approval for external tool delivery (Slack, email, etc.)
Artifact sent to approved destination
In-app viewing only β no external transmission
Gate-Speak
Phase 2Consent valid + speech enabled + no interrupt window + evidence available
MARIA speaks β rate-limited, evidence-backed response
Falls back to chat message or Live UI display only
Meetings are the missing input to Decision OS
This is not another meeting AI. This is how living decisions, commitments, and evidence flow into your governance pipeline β automatically.
Meetings are the last unstructured data
CRM, tickets, code β all structured. But 70% of critical decisions still live only in someone's memory of a meeting.
AI can execute, but can't judge without context
Autonomous agents need real decision history β not meeting summaries, but evidence-linked decision records to calibrate judgment.
Recording is not intelligence
Existing tools record and transcribe. That's raw material. Intelligence means structured decisions, tracked actions, and traceable evidence.
Remote-first made meetings the operating system
Your org runs on meetings. If meetings stay unstructured, your decision pipeline stays blind to 80% of actual operations.
Invite. Join. Done.
No SDK. No integration. No training. One calendar invite is all it takes.
Add MARIA to your Google Meet calendar invite
MARIA enters silently, announces presence, requests consent
Live minutes, decisions, and actions β immediately in your Decision OS