Meeting Agent · AI-Powered Meeting Intelligence

Architecture

One integration. Every platform.

Recall.ai handles the hard part — bot deployment across 6 platforms. 2nth.ai handles the intelligence — real-time processing, structured output, downstream routing.

Meeting

Teams · Meet
Zoom · Webex
Slack · GoTo

→

Recall.ai

Bot infra
Media capture
WebSocket stream

→

2nth.ai Worker
Transcript → AI
Summary · Actions
Routing logic

→

D1 + R2

Record storage
POPIA-compliant
Retention policy

→

Downstream

CRM · EHR
Tasks · Slack
Webhook · Email

🤖

Bot joins — you don't lift a finger

Via calendar integration or API trigger. Bot appears as a named participant — your brand, not ours. Announces itself and logs consent before a single word is captured.

Zero config per meeting

⚡

Real-time transcript, speaker-identified

Every utterance tagged with speaker name, timestamp, and confidence score. Live summary posted to chat on demand. Nothing lost, nothing misattributed.

Live WebSocket stream

📤

Structured output routed automatically

Within 60 seconds of meeting end: summary, decisions, action items with owners, open questions — delivered to CRM, EHR, task tool, Slack, or webhook. No copy-paste, no follow-up chase.

60s post-meeting delivery

Live meeting demo

The agent in the room.

A telehealth consultation — Erica Jankovich (dietitian) with a patient. Watch the meeting agent transcribe, detect key moments, and build the post-session summary in real time.

EJ

Erica Jankovich

🎙️ speaking

TM

Thandi M.

patient

🤖

EatFit AI Notetaker

transcribing

08:14

Erica

How has the bloating been since we moved off the high-FODMAP foods?

08:21

Thandi

Much better honestly — maybe 3 out of 10 now instead of the 7 it was at the start.

08:34

Erica

That's excellent progress. Any incidents this week? The cream pasta situation was last week right?

08:41

Thandi

Yes that was last week. This week was clean, I was much more careful at restaurants.

⚡ Live intelligence

✓ Bloating improvement detected: 7→3/10

! Inadvertent FODMAP exposure logged (cream pasta, last week)

→ Adherence: clean this week per patient report

📊 Meeting stats

Speaking time Erica 62% · Thandi 38%

Words transcribed 312

Confidence score 96.4%

Platform Google Meet · via Recall.ai

🔒 Compliance status

✓Consent announced & logged 00:00:08

✓No objections from 2 participants

✓Health data — POPIA S26 consent active

✓Retention: 30 days · SA region only

✓Recording to: imbila-deliverables R2

Industry routing examples

Same bot. Different output.

The meeting agent adapts its structured output to the industry. A legal firm gets verbatim minutes. A sales team gets CRM fields. A recruiter gets a scorecard. One skill — configured by the partner.

Configuration

Summary style: SOAP format (Subjective, Objective, Assessment, Plan)

Consent mode: POPIA Section 26 — health data, explicit consent required

Downstream: Practitioner copilot skill → EHR record

Retention: 6 years (HPCSA Health Professions Act requirement)

Special handling: Red flag detection (blood, weight loss, nocturnal symptoms)

⚠️ Scope enforcement

Meeting agent never generates diagnostic language in output. Clinical judgment fields in SOAP note are left blank for practitioner to complete. AI populates Subjective and Objective only — Assessment and Plan are practitioner responsibility.

"summary_style": "SOAP", "routed_to": [ "practitioner_copilot", "ehr_record", "booking_agent" ], "soap_note": { "subjective": "Patient reports bloating 3/10 (was 7/10). Clean adherence this week. One FODMAP exposure last week (cream pasta).", "objective": "Week 3 Low FODMAP elimination. Patient-reported improvement across all tracked metrics.", "assessment": "[PRACTITIONER TO COMPLETE]", "plan": "[PRACTITIONER TO COMPLETE]" }, "red_flags_detected": false, "retention_years": 6, "jurisdiction": "SA_POPIA_S26"

Configuration

Summary style: Verbatim minutes + decisions log + undertakings tracker

Consent mode: Legal privilege awareness — bot flags if meeting may be legally privileged

Downstream: Matter management system / DMS (NetDocuments, iManage)

Retention: Configurable per matter (7yr default)

Special handling: Undertakings extracted as action items with party and deadline

Legal privilege safeguard

If the calendar invite or participant list suggests attorney-client privilege, bot announces: "This meeting may be subject to legal privilege. By continuing, all parties consent to recording." If meeting is internal-only (attorney + client), bot can be configured to store locally without any cloud transmission.

"summary_style": "legal_minutes", "routed_to": ["matter_management"], "minutes": { "attendees": ["Adv. N. Dlamini", "S. van der Berg (client)"], "decisions": [ "Instruction accepted to oppose liquidation application", "Hearing date: 14 May 2026" ], "undertakings": [ { "party": "Adv. Dlamini", "task": "File opposing papers by 28 April", "deadline": "2026-04-28" } ] }, "privilege_flag": true, "storage": "local_only"

Configuration

Summary style: Sales brief — pain points, objections, next steps, deal signals

Downstream: Salesforce / HubSpot opportunity auto-update

CRM fields extracted: Stage, next action, decision maker, budget signals, timeline, competitors mentioned

Sentiment detection: Positive / neutral / concern / objection flagged per speaker

Coaching: Post-call rep coaching note (talk ratio, question rate, filler words)

Rep coaching note

Talk ratio: 68% (target <55%). Asked 3 discovery questions. "Um/uh" count: 14. Strong on product knowledge, missed buying signal at 14:32 — customer said "we'd need this before end of quarter." Suggest following up on timeline in next touchpoint.

"summary_style": "sales_brief", "routed_to": ["salesforce_opportunity"], "crm_update": { "stage": "Evaluation", "decision_maker": "CFO (not on call)", "budget_signal": "mentioned R180k budget ceiling", "timeline": "end of Q2 2026", "next_step": "Send proposal by Friday", "competitors": ["Salesforce"], "objections": ["Integration complexity with existing ERP"] }, "sentiment": "positive_with_concerns", "talk_ratio_rep": 0.68

Configuration

Summary style: Interview scorecard — competency ratings, red flags, hire recommendation

Downstream: ATS (Greenhouse, Lever, Workday)

Bias guardrail: Flags if interviewer comments reference protected characteristics (age, gender, ethnicity, disability) — does not include in output

Compliance: EEA (SA Employment Equity Act) aware — no discriminatory language recorded

Bias detection active

Any interviewer comment referencing protected characteristics under the Employment Equity Act is flagged and excluded from the scorecard. A compliance alert is sent to the HR admin. Interview recording access is restricted to hiring manager + HR only.

"summary_style": "interview_scorecard", "routed_to": ["greenhouse_ats"], "scorecard": { "candidate": "Sipho M.", "role": "Senior AI Engineer", "competencies": { "technical_depth": 4, "communication": 5, "problem_solving": 4, "culture_add": 3 }, "recommendation": "Advance to technical round", "red_flags": [], "bias_flags": [] }, "eea_compliant": true

Configuration

Summary style: Coaching session brief — insights, commitments, patterns, themes

Downstream: Coach dashboard, optional client portal share

Longitudinal: Tracks themes across sessions — "leadership identity" mentioned 4 sessions in a row

Confidentiality: Coach-client privilege mode — client must explicitly consent to any sharing

Longitudinal theme tracking

Across 6 sessions with this client: "imposter syndrome" mentioned in sessions 1, 3, 4, 6. "Delegation" theme emerging from session 4. Agent surfaces these patterns at the start of each new session so the coach can decide whether to address them explicitly.

"summary_style": "coaching_brief", "routed_to": ["coach_dashboard"], "session_brief": { "key_insight": "Client expressed readiness to step into MD role — first time this has been stated positively vs. with anxiety", "commitments": [ "Schedule difficult conversation with ops director this week", "Delegate Q2 planning to team leads" ], "recurring_themes": [ "imposter syndrome (sessions 1,3,4,6)", "delegation (sessions 4,5,6 — improving)" ] }, "confidentiality": "coach_only"

Compliance & safety

Non-negotiable before the first word is recorded.

Recording meetings without proper consent is illegal in most jurisdictions and exposes your practice and clients to significant risk. These are not features you turn on — they're defaults you can only make stricter.

✓

Consent — always announced

Bot announces itself within 8 seconds of joining. Configurable message. Consent event logged with timestamp and participant count. Bot leaves if host removes it — no silent recording.

✓

Data never leaves your region

Transcripts and recordings stored in your own R2 bucket in your configured region. Recall.ai processes media in real time but does not retain it beyond the API response window. No third-party training on your data.

✓

Retention policy enforced

You set the retention window (7 days / 30 days / 90 days / custom). Data is hard-deleted on schedule — not just hidden. Audit log maintained for the deletion event.

!

POPIA Section 26 — health data

Health meeting content (telehealth, HR medical) is special personal information. Explicit consent required — separate from general recording consent. Flagged automatically when healthcare practitioner account type is set.

!

Legal privilege detection

If calendar invite or participant list flags a legal matter, bot enters heightened consent mode and offers local-only storage (no cloud transmission) for attorney-client privilege protection.

🔒

Access control per meeting

Meeting records are scoped to the host's account by default. Sub-practice sharing requires explicit authorisation. All access logged. Partner admin can revoke access and trigger data deletion at any time.

Jurisdiction mode

Jurisdiction	Regulation	Consent standard	Bot behaviour
South Africa	POPIA · ECTA · HPCSA (health)	Legitimate interest (internal) or explicit consent (client-facing)	Announce + 10s opt-out window. Health data: explicit consent gate.
European Union	GDPR Art. 6 + Art. 9 (sensitive)	Explicit consent required for all participants	Consent request sent to each participant before bot joins. No join without response.
United States	ECPA + state wiretapping laws (CA, IL, FL: two-party)	One-party (most states) or two-party (CA, IL, FL)	Two-party mode: bot requests explicit acknowledgment from all participants in flagged states.
United Kingdom	UK GDPR · Investigatory Powers Act	Legitimate interest + transparency	Announce + data subject rights notice. Privacy notice URL posted to meeting chat.

Partner deployment

Your brand. Our infrastructure. Live in a day.

Partners configure and deploy the meeting agent under their own brand. The bot appears as "YourPractice AI Notetaker" — not 2nth.ai.

Bot display name EatFit AI Notetaker

Platforms enabled Meet · Teams · Zoom

Join mode

Summary style

Downstream routing

Data retention

Jurisdiction

Real-time summary

Rep coaching notes

POPIA S26 mode

What partners need to go live

1. Recall.ai API key

Sign up at recall.ai — $0.50/hr pay-as-you-go or custom plan. API key added to partner's 2nth.ai worker secrets. That's the only external dependency.

15 min setup

2. Calendar integration (optional)

Connect Google Calendar or Microsoft 365 via Recall.ai Calendar API. Bot auto-joins any meeting with a video link in the invite. No manual inviting required.

Auto-join enabled

3. Configure downstream routing

Choose where structured output goes — existing skills (practitioner copilot, booking agent), a webhook, email, or Slack. One config, applies to all meetings.

Works with existing skills

4. POPIA / compliance review

Review jurisdiction settings with your information officer. The consent flow is built in — you just need to confirm the lawful basis and retention period appropriate for your practice.

Non-negotiable step

Get Recall.ai API → Talk to Craig at 2nth.ai

Your AI agentis already in the room.

One integration. Every platform.

The agent in the room.

Same bot. Different output.

Non-negotiable before the first word is recorded.

Jurisdiction mode

Your brand. Our infrastructure. Live in a day.

What partners need to go live

Your AI agent
is already in the room.