Welcome to Speakr¶

Latest Release: v0.8.16-alpha - Per-Recording Model Selection & Backlog Cleanup

Per-upload model selection, embedding model env var, Mistral chunking, API v1 parity, and bug fixes

• Per-Upload / Per-Tag / Per-Folder Transcription Model - Configurable model dropdown via TRANSCRIPTION_MODELS_AVAILABLE
• Configurable Embedding Model - EMBEDDING_MODEL for local sentence-transformers, or EMBEDDING_BASE_URL to offload to an OpenAI-compatible provider
• Mistral Voxtral Chunking - MISTRAL_ENABLE_CHUNKING=true for long meeting recordings
• API v1 Parity - audio_duration, durations, folder, events, and deletion_exempt now exposed
• Bug Fixes - Reprocess now respects tag/folder hotwords; legacy "français" language values normalized; Cyrillic title generation no longer leaks unicode escapes

Version 0.8.19-alpha - Inquire-mode performance and re-embed reliability

Patch release on top of v0.8.18-alpha. No new features, no breaking changes.

• Vectorised chunk similarity search. Per-query Inquire search drops from 13-20s to under 1s on a ~17k-chunk library. A 4-enriched-query inquire turn now completes in 2-3s instead of ~60s.
• Embedding API retries on transient errors (rate limits, timeouts, 5xx, connection blips) with exponential backoff. Tunable via EMBEDDING_API_MAX_RETRIES and EMBEDDING_API_BACKOFF_SECONDS.
• process_recording_chunks rolls back when generate_embeddings returns fewer vectors than expected, preserving existing chunks instead of silently deleting them. Fixes the silent partial-failure mode that produced stale chunks across embedding-model swaps.
• Re-embed all loop has retry passes (default 2) and now picks up any recording with chunks in the table, regardless of current status.

Version 0.8.18-alpha - API v1 folder operations

Patch release on top of v0.8.17-alpha. Closes the gap from #274 follow-up where folder CRUD shipped in v0.8.16-alpha but recordings could not actually be moved between folders via the API.

• GET /api/v1/recordings?folder_id=<id> (or ?folder_id=none) filters list responses by folder
• PATCH /api/v1/recordings/{id} accepts folder_id (null to remove from any folder)
• PATCH /api/v1/recordings/batch accepts folder_id inside updates
• OpenAPI schema documents these plus previously-undocumented batch fields

Version 0.8.17-alpha - Bug fixes and CI maintenance

Patch release on top of v0.8.16-alpha. No new features, no breaking changes.

• Reprocess summary modal: prompt-variables panel and Append/Replace toggle now reflect the prompt source the user picked, rather than the recording's original tag
• Docs: corrected reverse-proxy nginx example (Connection header forwarded conditionally to avoid 500s on uploads through the proxy with Gunicorn). New Nginx Proxy Manager section.
• CI: bumped all GitHub Actions to Node 24 versions to clear deprecation warnings

Version 0.8.16-alpha - Prompt Templating, Transcription UX, and Observability

Prompt templating and summary control

• Prompt Template Variables - {{name}} placeholders in tag, folder, user-default, and admin-default summary prompts; values entered on the upload form, stored on the recording, and editable via reprocess
• Append vs Replace Mode - Reprocess summary modal and the new Customise summary prompt modal each let you Append to or Replace the resolved prompt
• Customise Summary Prompt Split-Button - A control next to Generate Summary opens the Append/Replace modal for recordings without a summary yet
• Full LLM Prompt Structure Preview - Admin and account pages show the complete two-message payload with chip-coded placeholders (blue for system tokens, amber for user variables); the user-side preview re-renders live as you type

Per-recording transcription control

• Per-Upload / Per-Tag / Per-Folder Transcription Model - TRANSCRIPTION_MODELS_AVAILABLE exposes a model dropdown across upload, reprocess, and tag/folder edit forms
• Admin-Managed Model List - Connectors with /v1/models discovery let admins curate the list from the dashboard
• Per-Connector Capability Gating - Hotwords, initial-prompt, and speaker-count UI hidden for connectors that don't support them
• Mistral Voxtral Chunking - MISTRAL_ENABLE_CHUNKING=true opts into app-side chunking for long recordings

ASR transcript editor

• Autosave - Saves 2s after the last keystroke when opted in
• Save Without Closing + Ctrl+S - New button keeps the editor open; Ctrl+S triggers save from anywhere in the editor
• Scroll Memory - Reopening the editor restores the previous scroll position
• Double-Click to Edit - Double-clicking a transcript row jumps into the editor with that segment highlighted

Account preferences

• Preferences Tab - New tab (split from Languages) with two-column layout for transcript display, editor behaviour, and language preferences
• Compact Timestamps - Opt-in mm:ss timestamps as two-part pill with speaker label; "Start" label on the leading segment
• Persist Recording-List Sort - Created/Meeting date toggle persists across sessions per browser (#263)

Embeddings and inquire mode

• Configurable Embedding Model - EMBEDDING_MODEL for local models
• API-Mode Embeddings - EMBEDDING_BASE_URL, EMBEDDING_API_KEY, EMBEDDING_DIMENSIONS route embeddings through any OpenAI-compatible provider
• Embedding Token Tracking + Re-Embed-All - Vector Store tab tracks embedding API cost separately and exposes a re-embed action

Observability and admin

• Per-Operation Token Stats - Title, summary, chat, event extraction, and embeddings as separate cards and charts
• Granular Token Budgets - TITLE_MAX_TOKENS and EVENT_MAX_TOKENS join the existing SUMMARY_MAX_TOKENS / CHAT_MAX_TOKENS
• LLM Timeout Visibility - Configured timeout logged at startup; APITimeoutError log entries include elapsed time

API v1

• Folder CRUD + Connector Discovery endpoints
• Recording Field Parity - audio_duration, durations, folder, events, deletion_exempt, prompt_variables, transcription model
• Forwarded Overrides - /api/v1/transcribe forwards transcription_model, hotwords, initial_prompt; custom-ASR connector forwards ?model= for WhisperX runtime switching

Bug fixes

• Reprocess now applies tag/folder/user default hotwords + initial_prompt (#265)
• Legacy transcription_language="français" normalised to ISO 639-1 on upgrade (#256)
• Title generation no longer leaks \\uXXXX escapes for non-ASCII transcripts (#260)
• CSRF token on Preferences form

Docs - prompt template variables guide; per-upload/tag/folder model selection; EMBEDDING_BASE_URL API mode across inquire-mode/vector-store/troubleshooting; nginx large-upload settings; Google Gemini OpenAI-compatible setup

Version 0.8.15-alpha - New Transcription Connectors & Upload API

• Mistral/Voxtral Connector - Cloud-based transcription with built-in speaker diarization via Mistral's Voxtral models
• VibeVoice Connector - Self-hosted transcription via vLLM with speaker diarization and automatic chunking
• Upload API: title & meeting_date - Optional metadata fields for integrations
• Regenerate Title - Button to re-generate recording title with AI
• Default Transcription Language - User-configurable default that auto-fills forms
• Tag-Driven Auto-Processing - Watch folders auto-apply tags and trigger via API
• Bug Fixes - Azure inquire crash, chat API serialization, user deletion, chunking limits

Version 0.8.13-alpha - Video Retention Fix

• Fixed large video files silently losing their video stream during upload when VIDEO_RETENTION=true
• Probe timeout now scales with file size; falls back to extension-based detection if probing fails

Version 0.8.12-alpha - Speaker Search, Shared Page Improvements & Bug Fixes

• Speaker Name Filter - Search/filter speakers by name on the Speakers Management page with bulk operation support
• Shared Page Auto-scroll - Follow-along mode on public shared transcript pages, matching the main app experience
• Dynamic Footer Year - Shared page footer year no longer hardcoded
• Duplicate Detection - SHA-256 file hashing with warning toasts and clickable copies indicator in sidebar/header
• Volume Controls - Volume slider popups and mute visual indicators on all audio/video players
• Speaker Enhancements - Split button UI, apply suggested names, name sanitization, JSON schema option, new speaker API endpoints
• Localization - Complete translations for folders, API tokens, recording recovery, events, and speakers management
• Bug Fixes - ASR_DIARIZE=false ignored, bulk delete cascade, file monitor stability, speaker snippets, null transcription, clean_llm_response too aggressive

Version 0.8.7 - Export Templates & Localization

• Customizable Export Templates - Create markdown templates for exports with variables ({{title}}, {{summary}}, {{notes}}) and conditionals for optional sections
• Localized Labels - Use {{label.metadata}}, {{label.summary}} etc. for automatically translated labels based on user's UI language
• Localized Dates - Export dates formatted per user's language preference (e.g., "15. Januar 2026" for German)
• Improvements - Opt-in ASR chunking, speaker ID remapping, ASR validation fixes

Fully backwards compatible with v0.8.x.

Version 0.8.3 - Naming Templates

• Custom Title Formatting - Create templates with variables ({{ai_title}}, {{filename}}, {{date}}) and regex patterns to extract data from filenames
• Tag or User Default - Assign templates to tags or set a user-wide default; templates without {{ai_title}} skip the AI call to save tokens
• API v1 Upload - New /api/v1/upload endpoint for programmatic recording uploads
• Improvements - Tag drag-and-drop reordering, registration domain restriction, event delete button, WebM seeking fix

Fully backwards compatible with v0.8.x.

Version 0.8.2 - Transcription Usage Tracking

• Transcription Budget Management - Set monthly transcription limits (in minutes) per user with 80% warnings and 100% blocking
• Usage Statistics - Track transcription minutes and estimated costs across all connectors
• Admin Dashboard Improvements - Redesigned stats layout with summary cards and per-user tables

Version 0.8.0 - Connector Architecture & REST API

• Connector-Based Transcription - Modular architecture with auto-detection for transcription providers
• OpenAI Diarization - Use gpt-4o-transcribe-diarize for speaker identification without self-hosting
• REST API v1 - Complete API for automation tools with Swagger UI at /api/v1/docs

See the Migration Guide and API Reference.

Version 0.6.6 - Filter & Compress

New Features - Audio compression and enhanced filtering

• Auto Compression - Lossless uploads automatically compressed (configurable codec/bitrate)
• Speaker Filtering - Filter recordings by speaker, starred/inbox toggles
• Sorting Fix - Sort toggle works correctly, added Upcoming group for future dates
• Format Support - .weba format, FFmpeg fallback for unknown formats

✅ Fully backward compatible. Optional env vars: AUDIO_COMPRESS_UPLOADS, AUDIO_CODEC, AUDIO_BITRATE

Version 0.6.5 - Separate Chat Model Configuration

New Feature - Configure different AI models for chat vs background tasks

• Separate Chat Model - Use different service tiers for chat and summarization (#143)
• Custom Datetime Picker - New themed calendar and time selection modal
• Bug Fixes - Audio chunking after refactor (#140), username display (#138)

✅ Fully backward compatible. Optional CHAT_MODEL_* environment variables.