Fireflies Calls Skill Skill

Fireflies Calls Skill

Search and analyze call recordings from Fireflies.ai with full transcript and AI-generated summaries.

When to Use This Skill

Use this skill when:

Find calls by keyword, title, or participant
Get call transcripts and summaries
Extract action items from meetings
Analyze conversations from specific dates
Review meeting notes and decisions

Quick Start Checklist

When user wants to find or analyze calls:

[ ] 1. Identify search criteria: keyword, participant, or date range
[ ] 2. Use FirefliesClient.search_meetings() with appropriate filters
[ ] 3. Review search results (duration now automatically accurate!)
[ ] 4. Get full details with get_meeting() if transcript needed
[ ] 5. Extract action items with get_action_items() if requested
[ ] 6. Format output with participant list and key insights

5-Second Decision Tree:

User mentions specific topic/keyword? → search_meetings(keyword="...")
User mentions person's name? → find_calls_with_person(person_name="...") (PREFERRED — checks transcript speakers, not just participants)
User asks about date/today? → search_meetings(date_from="...")

CRITICAL: Fireflies participants field is often EMPTY for Google Meet calls. Never rely solely on participant_email or participants list. Always use find_calls_with_person() when searching by person — it checks actual speaker names in transcripts.

Practical Workflow

BEFORE searching for calls:

Determine search type (keyword vs participant vs date)
Build search query:
- Topic → client.search_meetings(keyword="ECM", date_from="2025-11-01")
- Person → client.search_meetings(participant_email="contact@example.com")
- Recent → client.search_meetings(date_from=today, limit=50)
Get details if user needs transcript or action items
Format output with duration, participants, summary

Example rapid application:

User: "Find my calls about token launch this week"

Agent thinks:
- Keyword search needed ("token launch")
- Recent timeframe (this week)
- Use: client.search_meetings(keyword="token launch", date_from="2025-11-04")
- Duration automatically accurate (no API bug workaround needed)

Authentication

API Key: Automatically loaded from .env file:

FIREFLIES_API_KEY="your-api-key-here"

API URL: https://api.fireflies.ai/graphql

IMPORTANT: API key must be set in .env file. Never hardcode it in scripts.

Core Operations

1. Search Calls (Auto Real Duration!)

✅ FIXED: Client now automatically calculates accurate duration from transcripts by default! Fireflies API bug (wrong duration ~30-80 sec) is automatically corrected.

Quick search by keyword:

from data_sources.fireflies.fireflies_client import FirefliesClient
from datetime import datetime, timedelta

client = FirefliesClient()

# Search by keyword - duration is automatically accurate!
meetings = client.search_meetings(
    keyword="ECM",
    date_from=(datetime.now() - timedelta(days=7)).strftime("%Y-%m-%d"),
    limit=20
    # calculate_real_duration=True by default ✅
)

print(f"Found {len(meetings)} calls:\n")

for i, meeting in enumerate(meetings, 1):
    title = meeting.get('title', 'Untitled')
    date_ms = meeting.get('date', 0)
    if date_ms:
        date_obj = datetime.fromtimestamp(date_ms / 1000)
        date_str = date_obj.strftime('%Y-%m-%d %H:%M')
    else:
        date_str = 'Unknown'

    # Duration now accurate (in seconds)
    duration_sec = meeting.get('duration', 0)
    duration_min = duration_sec / 60

    participants = meeting.get('participants', [])
    meeting_id = meeting.get('id')

    print(f"{i}. {title}")
    print(f"   Date: {date_str}")
    print(f"   Duration: {duration_min:.1f} min")
    print(f"   Participants: {', '.join(participants[:3])}")
    print(f"   ID: {meeting_id}")
    print()

Search by participant email:

# Find all calls with specific person
meetings = client.search_meetings(
    participant_email="contact@example.com",
    date_from="2025-11-01",
    limit=50
)

Combined search:

# Keyword + participant + date range
meetings = client.search_meetings(
    keyword="token launch",
    participant_email="rakhan@example.com",
    date_from="2025-11-01",
    date_to="2025-11-04",
    limit=20
)

1b. Find Calls by Person (Speaker Detection)

PREFERRED method when searching by person name. Checks transcript speakers, not just participants.

from data_sources.fireflies.fireflies_client import FirefliesClient
from datetime import datetime

client = FirefliesClient()
today = datetime.now().strftime('%Y-%m-%d')

# Find all calls where Nick Snopov spoke today
meetings = client.find_calls_with_person(
    person_name="Nick Snopov",  # case-insensitive, partial match OK
    date_from=today
)

for m in meetings:
    title = m.get('title', 'Untitled')
    speakers = m.get('speakers', [])
    duration_min = m.get('duration', 0) / 60
    match_src = m.get('match_source', '?')
    print(f"{title} | {duration_min:.1f} min | matched via: {match_src}")
    print(f"  Speakers: {speakers}")

Why this exists: Fireflies API leaves participants empty for ~60% of Google Meet calls. The search_meetings(participant_email=...) will miss these calls entirely. find_calls_with_person() scans actual speaker_name fields in transcripts — never misses.

2. Get Full Call Details with Transcript

Complete meeting analysis:

# Get meeting ID from search results
meeting_id = "01K8RH3QTPMF92ZR63K0TK516D"

# Get full details including transcript
meeting = client.get_meeting(meeting_id, include_transcript=True)

# Display formatted output
print("=" * 100)
print(f"📅 {meeting.get('title')}")
print(f"ID: {meeting_id}")

# Date
date_ms = meeting.get('date', 0)
if date_ms:
    date_obj = datetime.fromtimestamp(date_ms / 1000)
    print(f"Date: {date_obj.strftime('%A, %B %d, %Y at %H:%M')}")

# Duration
duration = meeting.get('duration', 0)
if duration:
    print(f"Duration: {duration / 60:.1f} minutes")

# Participants
participants = meeting.get('participants', [])
if participants:
    print(f"\n👥 Participants ({len(participants)}):")
    for p in participants:
        print(f"   - {p}")

# Summary
summary = meeting.get('summary', {})
if summary:
    print("\n📝 SUMMARY:")
    print("-" * 100)

    if summary.get('overview'):
        print(f"\nOverview:\n{summary['overview']}")

    if summary.get('keywords'):
        print(f"\nKeywords: {', '.join(summary['keywords'])}")

    if summary.get('action_items'):
        print(f"\n✅ Action Items:\n{summary['action_items']}")

# Transcript
sentences = meeting.get('sentences', [])
if sentences:
    print(f"\n💬 TRANSCRIPT ({len(sentences)} sentences):")
    print("-" * 100 + "\n")

    current_speaker = None
    for sentence in sentences[:20]:  # First 20 sentences
        speaker = sentence.get('speaker_name', 'Unknown')
        text = sentence.get('text', '')

        if speaker != current_speaker:
            print(f"\n{speaker}:")
            current_speaker = speaker

        print(f"  {text}")

3. Extract Action Items

Get only action items:

meeting_id = "01K8RH3QTPMF92ZR63K0TK516D"

actions = client.get_action_items(meeting_id)

if actions:
    print("✅ Action Items:")
    for i, action in enumerate(actions, 1):
        print(f"{i}. {action}")
else:
    print("No action items found")

Common Search Patterns

Pattern 1: Get Today's Calls

from datetime import datetime

# TODAY's calls - duration automatically accurate!
today = datetime.now().strftime('%Y-%m-%d')

meetings = client.search_meetings(
    date_from=today,
    limit=50
    # Real duration calculated automatically ✅
)

print(f"📅 Today's calls: {len(meetings)}\n")

total_duration = 0
for i, meeting in enumerate(meetings, 1):
    date_ms = meeting.get('date', 0)
    time_str = datetime.fromtimestamp(date_ms / 1000).strftime('%H:%M')

    duration_sec = meeting.get('duration', 0)
    duration_min = duration_sec / 60
    total_duration += duration_sec

    title = meeting.get('title', 'Untitled')
    participants = meeting.get('participants', [])

    print(f"{i}. [{time_str}] {title}")
    print(f"   Duration: {duration_min:.1f} min")
    print(f"   Participants: {len(participants)} people")
    print()

print(f"\n📊 Total: {total_duration/60:.1f} minutes ({total_duration/3600:.1f} hours)")

Pattern 2: Find Calls About Specific Topic

# Search by keywords
keywords = ["token launch", "ECM", "bio.xyz", "funding"]

for keyword in keywords:
    meetings = client.search_meetings(keyword=keyword, limit=10)
    print(f"\n🔍 '{keyword}': {len(meetings)} calls")

    for meeting in meetings[:3]:  # Top 3
        print(f"   - {meeting.get('title')} ({meeting.get('id')})")

Pattern 3: Participant Call History

# All calls with specific person
email = "contact@example.com"

meetings = client.search_meetings(
    participant_email=email,
    date_from="2025-01-01",
    limit=50
)

print(f"📊 Call history for {email}:")
print(f"Total calls: {len(meetings)}\n")

for meeting in meetings:
    date_ms = meeting.get('date', 0)
    if date_ms:
        date_str = datetime.fromtimestamp(date_ms / 1000).strftime('%Y-%m-%d')
    else:
        date_str = 'Unknown'

    print(f"   {date_str} - {meeting.get('title')}")

Pattern 4: Summary Only (No Transcript)

# Faster if you don't need full transcript
meeting = client.get_meeting(meeting_id, include_transcript=False)

summary = meeting.get('summary', {})
print("Quick Summary:")
print(f"Overview: {summary.get('overview', 'N/A')}")
print(f"Keywords: {', '.join(summary.get('keywords', []))}")

API Methods Reference

find_calls_with_person()

PREFERRED for person search. Checks transcript speakers, not just participants.

Parameters:

person_name (str): Name to search (case-insensitive, partial match). E.g. "Nick Snopov", "Nikita"
date_from (str): Start date (YYYY-MM-DD)
date_to (str): End date (YYYY-MM-DD)
limit (int): Max meetings to scan (default: 50)

Returns: List of meetings where the person was a speaker, with extra fields:

speakers - List of actual speaker names from transcript
match_source - "participants" or "transcript_speakers"

Why it exists: Fireflies leaves participants empty for ~60% of Google Meet calls. search_meetings(participant_email=...) will miss them. This method scans speaker_name in transcripts — never misses.

search_meetings()

Parameters:

keyword (str): Search in transcript content
title (str): Search in meeting title
participant_email (str): Filter by participant
organizer_email (str): Filter by organizer
host_email (str): Filter by host
date_from (str): Start date (YYYY-MM-DD or ISO)
date_to (str): End date (YYYY-MM-DD or ISO)
limit (int): Max results (default: 50, max: 50)
skip (int): Pagination offset (default: 0)
calculate_real_duration (bool): Calculate accurate duration from transcript (default: True ✅)
- Set to False only for faster queries when duration not critical

⚠️ API BUG FIXED: Fireflies API returns wrong duration (~30-80 sec). Client now auto-corrects this by default!

Returns: List of meetings with:

id - Meeting ID
title - Meeting title
date - Unix timestamp (milliseconds)
duration - Duration in seconds (automatically accurate from transcript)
real_duration - Real duration from transcript (same as duration when auto-calculated)
participants - List of emails
organizer_email - Organizer email
audio_url - Audio recording URL
video_url - Video recording URL
summary - AI-generated summary object

get_meeting()

Parameters:

meeting_id (str): Fireflies meeting ID
include_transcript (bool): Include full transcript (default: True)

Returns: Complete meeting data including:

All fields from search_meetings()
sentences - Full transcript (if include_transcript=True)
- text - Sentence text
- speaker_name - Speaker name
- start_time - Start time in seconds
- end_time - End time in seconds

get_action_items()

Parameters:

meeting_id (str): Fireflies meeting ID

Returns: List of action items as strings

get_user_info()

Parameters: None

Returns: User information:

email - User email
name - User name
minutes_consumed - Minutes of calls processed

Example: Complete Call Analysis Workflow

from data_sources.fireflies.fireflies_client import FirefliesClient
from datetime import datetime, timedelta

client = FirefliesClient()

# Step 1: Search for calls
print("🔍 Searching for calls about 'ECM'...")
date_from = (datetime.now() - timedelta(days=7)).strftime("%Y-%m-%d")

meetings = client.search_meetings(
    keyword="ECM",
    date_from=date_from,
    limit=20
)

if not meetings:
    print("No calls found")
    exit()

print(f"Found {len(meetings)} calls\n")

# Step 2: Select relevant call
for i, meeting in enumerate(meetings, 1):
    title = meeting.get('title', 'Untitled')
    print(f"{i}. {title} (ID: {meeting.get('id')})")

# Step 3: Get full details for first call
selected_meeting_id = meetings[0].get('id')
print(f"\n📥 Analyzing: {meetings[0].get('title')}\n")

meeting = client.get_meeting(selected_meeting_id, include_transcript=True)

# Step 4: Extract key information
summary = meeting.get('summary', {})
sentences = meeting.get('sentences', [])

print("=" * 100)
print(f"📅 {meeting.get('title')}")
print(f"Date: {datetime.fromtimestamp(meeting.get('date', 0) / 1000).strftime('%Y-%m-%d %H:%M')}")
print(f"Duration: {meeting.get('duration', 0) / 60:.1f} minutes")
print(f"Participants: {', '.join(meeting.get('participants', []))}")

print("\n📝 SUMMARY:")
print(summary.get('overview', 'N/A'))

print("\n✅ ACTION ITEMS:")
print(summary.get('action_items', 'N/A'))

print(f"\n💬 TRANSCRIPT ({len(sentences)} sentences):")
current_speaker = None
for sentence in sentences[:15]:
    speaker = sentence.get('speaker_name')
    if speaker != current_speaker:
        print(f"\n{speaker}:")
        current_speaker = speaker
    print(f"  {sentence.get('text')}")

print("\n" + "=" * 100)

Command Line Usage

The FirefliesClient also provides CLI:

# Search calls
python data_sources/fireflies/fireflies_client.py search --keyword "ECM" --from 2025-11-01 --limit 10

# Get meeting details
python data_sources/fireflies/fireflies_client.py get 01K8RH3QTPMF92ZR63K0TK516D

# Get action items only
python data_sources/fireflies/fireflies_client.py actions 01K8RH3QTPMF92ZR63K0TK516D

# Get user info
python data_sources/fireflies/fireflies_client.py user

Related Files

Client Implementation: data_sources/fireflies/fireflies_client.py
Example Usage: algorithms/A8_G&A_div/Daniel Personal/finetune_on_calls_emails_docs/01_fetch_last_call.py
CLAUDE.md Reference: Section 1.1¶6 - Fireflies operations

Version History

v1.0 (2025-01-15): Initial skill creation
- Search calls by keyword, participant, date
- Get full transcripts and summaries
- Extract action items
- CLI and Python API

Agent Skills: Fireflies Calls Skill

Install this agent skill to your local

Skill Files