Fireflies.ai

Automatically captures and transcribes audio from video calls (Zoom, Google Meet, Microsoft Teams, Slack) and phone conversations using speech-to-text APIs with speaker identification. The system integrates directly with calendar and meeting platforms to detect when calls begin, initiates recording with participant consent, and processes audio streams through multi-speaker diarization models to attribute spoken segments to individual participants, generating timestamped transcripts with speaker labels.

Processes completed transcripts through large language models to generate structured summaries that extract key decisions, action items with assigned owners, topics discussed, and sentiment. The system uses prompt engineering and fine-tuned models to identify action items with implicit ownership (e.g., 'we need to fix the database' → identifies engineer responsible), generates executive summaries at multiple detail levels (1-line, paragraph, bullet-point), and tags summaries by topic for organizational purposes.

Automatically detects and redacts personally identifiable information (PII), payment card data, and other sensitive information from transcripts before storage or sharing. The system uses NLP-based entity recognition to identify names, email addresses, phone numbers, credit card numbers, SSNs, and other sensitive data, then redacts or masks them in transcripts and summaries. Redaction is configurable per data type and can be applied retroactively to existing transcripts. Audit logs track what was redacted and when.

Integrates with calendar systems (Google Calendar, Outlook) to automatically detect meetings, extract attendee information, and provide pre-meeting context from previous conversations with the same participants. The system suggests optimal meeting times based on participant availability and past meeting patterns, provides meeting agendas generated from previous discussions with attendees, and sends pre-meeting briefings with relevant context from past calls. Post-meeting, it automatically updates calendar entries with summaries and action items.

Indexes all transcripts in a vector database using embeddings, enabling semantic search that finds relevant meetings based on meaning rather than keyword matching. Users can search for concepts ('discuss pricing strategy'), specific topics ('customer churn concerns'), or questions ('what did we decide about the API?'), and the system returns ranked results with highlighted relevant segments and timestamps. Search results include context snippets showing the relevant discussion with speaker attribution.

Aggregates data across multiple transcripts to identify patterns, recurring topics, sentiment trends, and conversation dynamics over time. The system analyzes speaker participation rates, topic frequency across meetings, sentiment evolution for specific customers or projects, and flags anomalies (e.g., sudden shift in customer tone, repeated unresolved issues). Results are presented as dashboards showing trends, heatmaps of topic frequency, and comparative metrics across teams or time periods.

Integrates with CRM systems (Salesforce, HubSpot, Pipedrive) and productivity tools (Slack, Notion, Asana) to automatically sync meeting summaries, action items, and insights. The system maps extracted action items to CRM deal records, posts meeting summaries to Slack channels, creates tasks in Asana with due dates and assignees, and updates contact records with call notes. Integration uses webhook-based event streaming and API polling to maintain bidirectional sync without manual data entry.

Allows teams to fine-tune Fireflies' transcription and summarization models on domain-specific vocabulary and jargon. Users can upload glossaries, past transcripts with corrections, or custom training data to improve accuracy for industry-specific terms (e.g., medical terminology, technical product names, legal concepts). The system retrains embedding and language models on this custom data, improving both transcription accuracy and summary relevance for specialized domains.

Provides enterprise-grade compliance features including audit logs of all transcript access, data retention policies with automatic deletion, encryption at rest and in transit, and compliance certifications (SOC 2, HIPAA, GDPR). The system tracks who accessed which transcripts, when, and for how long; enforces data retention windows (e.g., delete all recordings after 90 days); and provides compliance reports for audits. Encryption uses AES-256 for stored data and TLS 1.3 for transmission.

Analyzes sales calls using conversation intelligence metrics to score call quality, identify best practices, and flag coaching opportunities. The system evaluates talk-to-listen ratio, question-asking frequency, objection handling, discovery depth, and alignment with sales methodology (e.g., MEDDIC, Sandler). Scores are benchmarked against team averages and top performers, with detailed feedback on specific moments in the call where improvement is needed. Coaching recommendations are generated based on identified gaps.

Displays live transcription and AI-generated insights during active meetings, showing real-time speaker attribution, key topics being discussed, and action items as they emerge. The system uses streaming transcription APIs to process audio in near-real-time (2-3 second latency), identifies topics and action items incrementally as the conversation progresses, and surfaces them in a side panel or dedicated dashboard. Participants can see live summaries and action items without waiting for post-meeting processing.

Builds speaker profiles across multiple meetings, learning to identify and distinguish individual speakers even when they join different calls. The system uses voice biometrics and conversation patterns to recognize recurring speakers, maintains speaker profiles with metadata (name, role, company, contact info), and enables searching for all meetings involving specific people. Profiles are automatically enriched with CRM data when available, linking speakers to customer records or team members.