Call My Link

Captures video and audio streams from all meeting participants in real-time, encoding them into a unified media file with synchronized multi-track audio. The system likely uses WebRTC APIs to intercept media streams at the browser level, then muxes them into a container format (MP4/WebM) with metadata tagging for each participant's track, enabling later selective playback or transcription of individual speakers.

Converts recorded audio into searchable text transcripts while identifying and labeling which participant spoke each segment. The system likely sends audio to a cloud speech-to-text API (Google Cloud Speech-to-Text, Azure Speech Services, or Deepgram) and applies speaker diarization algorithms (clustering audio embeddings by speaker characteristics like pitch and timbre) to attribute segments to participants. Diarization may be seeded with participant metadata from the call to improve accuracy.

Generates concise summaries of transcribed calls by identifying and extracting key discussion points, decisions, and action items using extractive and abstractive summarization techniques. The system likely uses an LLM (GPT-4, Claude, or similar) with a prompt that instructs it to parse the transcript, identify semantic clusters (topics discussed), extract decisions and commitments, and generate a structured summary. May include post-processing to deduplicate action items and link them to responsible parties.

Generates unique, time-limited URLs that allow non-participants to view or listen to recorded calls without requiring them to log in or install software. The system implements a token-based access control layer where each link encodes permissions (view-only, download-allowed, expiration time) and is validated server-side before serving the media. Links likely use short URL generation (bit.ly-style) for easy sharing via email or chat, with optional password protection for sensitive calls.

Manages persistent storage of video and audio files with configurable retention policies, archival, and deletion workflows. The system likely stores recordings in cloud object storage (AWS S3, Google Cloud Storage, or Azure Blob) with metadata indexed in a database for search and retrieval. Lifecycle policies (e.g., auto-delete after 90 days, archive to cold storage after 30 days) are applied based on user tier or explicit configuration. Freemium tier likely has strict storage quotas (e.g., 2-5 GB) to encourage upgrades.

Enables full-text search across all transcribed calls and summaries using keyword matching and semantic search. The system likely indexes transcripts in a search engine (Elasticsearch, Algolia, or similar) with fields for speaker, timestamp, and summary content. Semantic search may use embeddings (stored in a vector database) to find conceptually similar calls even if keywords don't match. Search results return matching segments with context (surrounding sentences) and timestamps for easy navigation.

Automatically links recorded calls to calendar events and enables one-click recording start from calendar invites. The system likely uses OAuth to connect to Google Calendar, Outlook, or similar services, then matches recorded calls to calendar events by comparing timestamps and participant lists. May support pre-call setup where users can enable recording from the calendar invite, with the recording automatically associated with the event post-call.

Enables users to perform operations (transcribe, summarize, delete, export) on multiple calls simultaneously rather than one at a time. The system likely implements a job queue (Celery, Bull, or similar) that processes bulk requests asynchronously, with progress tracking and completion notifications. Bulk operations may be triggered via UI (checkbox select) or API (batch endpoint), with results aggregated and downloadable as a CSV or JSON file.

Allows meeting organizers to record calls while giving participants the option to opt out of being recorded (video/audio excluded from final recording). The system likely implements a pre-call consent screen or in-call toggle where participants can disable their camera/microphone from the recording without disconnecting from the call. The recording system then mutes or blanks their track in the final output, or excludes it entirely if they opt out before joining.