Imported from jeremylongshore/tons-of-skills-marketplace (skills/.curated/fireflies-reference-architecture/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill fireflies-reference-architecture. Copyright stays with the author (MIT).

Fireflies.ai Reference Architecture

Overview

Production architecture for meeting intelligence using Fireflies.ai. Event-driven pipeline: meetings are recorded by the Fireflies bot, transcripts arrive via webhook, then are processed for action items, analytics, and CRM sync.

Prerequisites

• A documented consent, access, retention, and downstream-sharing policy for each meeting source.
• Scoped identities separating webhook ingress, transcript processing, storage, and CRM consumers.
• A policy owner, synthetic fixtures, redacted telemetry, and an emergency disable/rollback path.

Instructions

1. Verify event signatures at ingress and place opaque IDs on a durable, idempotent queue.
2. Retrieve and process transcript data only within the authorized worker; apply field allowlists before any downstream sync.
3. Enforce retention and access checks at storage and retrieval, and emit aggregate/redacted operational metrics.
4. Test with fictional meetings, promote through a canary, and stop processing if consent, access, or destination validation fails.

Examples

Route a synthetic transcript-ready event through the queue using an opaque ID. The worker writes only approved action items to a staging CRM, rejects a repeated event as a duplicate, and records no transcript text in logs or metrics.

Architecture

┌──────────────────────────────────────────────────────┐
│              Meeting Sources                          │
│  Zoom  │  Google Meet  │  MS Teams  │  Upload API    │
└──────────┬───────────────────────────────┬───────────┘
           │ Bot auto-joins                │ uploadAudio
           ▼                               ▼
┌──────────────────────────────────────────────────────┐
│              Fireflies.ai Platform                    │
│  Transcription → Speaker ID → AI Summary → Actions   │
└───────────────────────┬──────────────────────────────┘
                        │ Webhook: "Transcription completed"
                        │ Payload: { meetingId, eventType }
                        ▼
┌──────────────────────────────────────────────────────┐
│              Your Webhook Receiver                    │
│  1. Verify x-hub-signature (HMAC-SHA256)             │
│  2. ACK 200 immediately                              │
│  3. Queue for async processing                       │
└───────────────────────┬──────────────────────────────┘
                        │
           ┌────────────┼────────────┐
           ▼            ▼            ▼
   ┌──────────────┐ ┌────────┐ ┌──────────────┐
   │ Transcript   │ │ Action │ │ Analytics    │
   │ Storage      │ │ Items  │ │ Engine       │
   │ (DB/Search)  │ │ (CRM)  │ │ (Dashboards) │
   └──────────────┘ └────────┘ └──────────────┘

Core Components

1. GraphQL Client Layer

// lib/fireflies.ts
const FIREFLIES_API = "https://api.fireflies.ai/graphql";

export async function firefliesQuery(query: string, variables?: any) {
  const res = await fetch(FIREFLIES_API, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${process.env.FIREFLIES_API_KEY}`,
    },
    body: JSON.stringify({ query, variables }),
  });
  const json = await res.json();
  if (json.errors) throw new Error(json.errors[0].message);
  return json.data;
}

2. Webhook Processor

// services/webhook-processor.ts
import crypto from "crypto";

interface TranscriptEvent {
  meetingId: string;
  eventType: string;
  clientReferenceId?: string;
}

export async function processWebhookEvent(event: TranscriptEvent) {
  // Fetch full transcript
  const { transcript } = await firefliesQuery(`
    query($id: String!) {
      transcript(id: $id) {
        id title date duration
        organizer_email
        speakers { id name }
        sentences {
          speaker_name text start_time end_time
          ai_filters { task question sentiment }
        }
        summary { overview action_items keywords topics_discussed }
        meeting_attendees { displayName email }
        analytics {
          sentiments { positive_pct negative_pct neutral_pct }
          speakers { name duration word_count questions words_per_minute }
        }
      }
    }
  `, { id: event.meetingId });

  // Process in parallel
  await Promise.all([
    storeTranscript(transcript),
    syncActionItems(transcript),
    updateAnalytics(transcript),
  ]);

  return transcript;
}

3. Transcript Storage

// services/transcript-store.ts
interface StoredMeeting {
  firefliesId: string;
  title: string;
  date: string;
  duration: number;
  speakers: string[];
  overview: string;
  actionItems: string[];
  keywords: string[];
  sentiment: { positive: number; negative: number; neutral: number };
}

async function storeTranscript(transcript: any): Promise<StoredMeeting> {
  const meeting: StoredMeeting = {
    firefliesId: transcript.id,
    title: transcript.title,
    date: transcript.date,
    duration: transcript.duration,
    speakers: transcript.speakers.map((s: any) => s.name),
    overview: transcript.summary?.overview || "",
    actionItems: transcript.summary?.action_items || [],
    keywords: transcript.summary?.keywords || [],
    sentiment: {
      positive: transcript.analytics?.sentiments?.positive_pct || 0,
      negative: transcript.analytics?.sentiments?.negative_pct || 0,
      neutral: transcript.analytics?.sentiments?.neutral_pct || 0,
    },
  };

  // Store in your database
  await db.meetings.upsert({ where: { firefliesId: meeting.firefliesId }, data: meeting });
  return meeting;
}

4. Action Item Sync

// services/action-items.ts
async function syncActionItems(transcript: any) {
  const items = transcript.summary?.action_items || [];
  if (items.length === 0) return;

  const attendees = transcript.meeting_attendees?.map((a: any) => a.email) || [];

  for (const item of items) {
    await taskManager.create({
      title: item.slice(0, 200),
      source: `Fireflies: ${transcript.title}`,
      meetingId: transcript.id,
      meetingDate: transcript.date,
      participants: attendees,
    });
  }

  console.log(`Synced ${items.length} action items from "${transcript.title}"`);
}

5. Meeting Analytics

// services/analytics.ts
async function buildWeeklyReport() {
  const since = new Date(Date.now() - 7 * 86400000).toISOString();

  const data = await firefliesQuery(`
    query($fromDate: DateTime) {
      transcripts(fromDate: $fromDate, limit: 100) {
        id title date duration
        participants
        summary { action_items keywords }
        analytics {
          speakers { name duration word_count }
          sentiments { positive_pct }
        }
      }
    }
  `, { fromDate: since });

  const meetings = data.transcripts;
  return {
    totalMeetings: meetings.length,
    totalHours: (meetings.reduce((s: number, m: any) => s + m.duration, 0) / 60).toFixed(1),
    actionItems: meetings.reduce((s: number, m: any) => s + (m.summary?.action_items?.length || 0), 0),
    topKeywords: aggregateKeywords(meetings).slice(0, 10),
    avgSentiment: avgSentiment(meetings),
  };
}

function aggregateKeywords(meetings: any[]): [string, number][] {
  const counts: Record<string, number> = {};
  for (const m of meetings) {
    for (const kw of m.summary?.keywords || []) {
      counts[kw] = (counts[kw] || 0) + 1;
    }
  }
  return Object.entries(counts).sort((a, b) => b[1] - a[1]);
}

6. Audio Upload Pipeline

// services/upload.ts
async function uploadRecording(fileUrl: string, title: string, attendees?: any[]) {
  return firefliesQuery(`
    mutation($input: AudioUploadInput) {
      uploadAudio(input: $input) {
        success
        title
        message
      }
    }
  `, {
    input: {
      url: fileUrl,
      title,
      attendees: attendees?.map(a => ({ displayName: a.name, email: a.email })),
      webhook: process.env.WEBHOOK_URL,
      client_reference_id: `upload-${Date.now()}`,
    },
  });
}

Project Layout

meeting-intelligence/
  src/
    lib/fireflies.ts          # GraphQL client
    services/
      webhook-processor.ts    # Event handler
      transcript-store.ts     # DB persistence
      action-items.ts         # CRM/task sync
      analytics.ts            # Aggregation
      upload.ts               # Audio upload
    api/
      webhooks/fireflies.ts   # Webhook endpoint
      health.ts               # Health check
    types/fireflies.ts        # TypeScript interfaces
  tests/
    fixtures/                 # Recorded API responses
    services/                 # Service unit tests

Error Handling

Output

• Event-driven architecture with webhook-triggered processing
• Transcript storage with search-ready schema
• Action item extraction and CRM sync pipeline
• Meeting analytics aggregation engine

Resources

• Fireflies API Docs
• Fireflies Webhooks
• Transcript Query

Next Steps

For multi-environment deployment, see fireflies-multi-env-setup.