Phase 1: Dead Code Cleanup
- Remove GeminiGroundedProvider import and property from linkedin_service.py
- Remove fallback_provider property (gemini_provider imports)
- Fix routers/linkedin.py edit endpoint to use llm_text_gen
- Delete dead LinkedInImageEditor class
- Remove dead _transform_gemini_sources from content_generator.py
Phase 2: Research Infrastructure Alignment
- Add user_id to _conduct_research() for pre-flight validation
- Add validate_exa_research_operations() before Exa/Tavily calls
- Pass user_id to provider.simple_search() for usage tracking
- Inject research content into LLM prompts via _build_research_context()
- Fix Google engine path to fallback to Exa
- Add Exa → Tavily fallback on research failure
Phase 3: Cosmetic Cleanup
- Rename _generate_prompts_with_gemini → _generate_prompts_with_llm
- Rename _build_gemini_prompt → _build_image_prompt
- Rename _parse_gemini_response → _parse_llm_response
- Remove all Gemini references from LinkedIn code (0 remaining)
- Update docstrings and log messages
Additional:
- Research caching using existing ResearchCache
- Shared ExaContentResearchProvider in services/research/
- Persona service uses llm_text_gen instead of gemini_structured_json_response
- LinkedInWriter.tsx ChatMessage → ChatMsg type mapping fix
- RegisterLinkedInActionsEnhanced.tsx content_format_rules typing fix
#3 — Duplicate prospect handling: add_lead now checks (campaign_id, url)
before insert; bulk_add_leads skips existing URLs.
#8 — Atomic rate limiting: try_increment_* methods atomically check cap
and increment in a single session; router uses these before send.
#10 — Reply matching via Message-ID: sender generates Message-ID header,
stored on OutreachAttempt; reply monitor parses In-Reply-To/References;
poll_replies matches by message_id first, falls back to from_email.
#11 — Save-to-campaign uses existing store results instead of
re-running expensive deepDiscover.
#12 — Lead status Literal type: Pydantic models enforce valid status
values; backend validates via LEAD_VALID_STATUSES frozenset;
frontend API typed as LeadStatus union.
- Add BacklinkOutreachScraper (Exa + DuckDuckGo deep scraping)
- Extend DB and Pydantic models for lead enrichment columns
- Add StorageService methods for lead CRUD with auto-migration
- Add backend endpoints: deep discover, campaign detail, lead management
- Extend frontend API client and store with discovery + lead actions
- Create BacklinkOutreachDashboard component with campaigns/discover/leads tabs
- Register route at /backlink-outreach under SEO feature flag
- Add nav entry under Enterprise & Advanced in tool categories
- Fix text selection menu not showing: wire contentRef via inputRef on multiline TextField
- Fix blog title not truncating: add min-w-0 for flex item overflow
- Fix outline generation 500: escape curly braces in f-string prompt template
- Fix content generation 'NoneType not callable': replace SessionLocal() with get_session_for_user(), add db param to MediumBlogGenerator, fix signature mismatch in database_task_manager
- Fix writing assistant suggest 500: add auth + user_id to API endpoint and service, replace sync requests with httpx.AsyncClient
- Fix hallucination detector 404: explicitly include router in main.py and app.py
- Fix missing error_data in task failure responses
- Hide CopilotKit web inspector button
- Remove hardcoded fallback suggestions from SmartTypingAssist
- Fix stale closure refs in SmartTypingAssist handleTypingChange
- Add two-column editor layout, stats bar, section hover menu
- Various subscription, billing, and research module improvements
Backend:
- product_image_service.py: Replaced direct wavespeed_client.generate_image()
with generate_image() from main_image_generation (unified entry point)
- This ensures subscription pre-flight validation (_validate_image_operation)
and usage tracking (_track_image_operation_usage) are enforced
- Removed _generate_image_with_retry method and WaveSpeedClient dependency
- Animation/video/avatar services already route through ImageStudioManager - no changes needed
Frontend:
- useProductMarketing.ts: Added formatError() helper for 402/429 detection
across all 8 API operations
- useCampaignCreator.ts: Added formatError() helper for 402/429 detection
across all 13 API operations
- All error messages now surface subscription limits with upgrade prompts
Changes:
1. helpers.py (_track_image_operation_usage): Map provider name to DB columns
dynamically (stability→stability_calls, wavespeed→wavespeed_calls, etc.)
instead of hardcoding stability_calls/stability_cost.
2. upscale_service.py: Added _track_image_operation_usage() call after
successful Stability upscale completion.
3. control_service.py: Added _track_image_operation_usage() call after
successful Stability control operation completion.
4. edit_service.py: Added _track_image_operation_usage() call after
successful Stability edit operation (remove_background, inpaint,
outpaint, search_replace, search_recolor, relight).
Previously only Create Studio and Face Swap tracked usage. Now all five
studios correctly decrement subscription limits.
Frontend Changes:
- Add scene numbering badge (1/N) next to scene titles
- Add inline status chips (Complete, Audio, Image, Voice, Why Script)
- Professional AI-like gradient styling for all chips with shadows
- Remove Script Editor header and 'Why This Script Format?' collapsible
- Move Voice and Why Script info to per-scene chips
- Make scene section mobile-responsive (responsive layout, button sizing)
- Rename 'B-Roll Charts' to 'Podcast Charts' with accordion (collapsed by default)
- Add sceneIndex prop to SceneEditor for scene numbering
- Enhanced accessibility with keyboard navigation and focus states
Backend Changes:
- Audio handler improvements
- B-roll handler enhancements
- Script handler updates
- B-roll composer and service improvements
- Removed temporary broll_temp files
Technical:
- Full mobile responsiveness for scene cards
- Gradient chip styling: vibrant colors with white text and shadows
- Non-breaking approval/generation flow preserved
- TypeScript compatibility maintained
- Fix voice clone preview saved as .wav regardless of actual format (MP3/WebM
content from WaveSpeed was saved with .wav extension causing NotSupportedError)
- Add detect_audio_format() and ensure_audio_extension() to media_utils
- Fix assets_serving.py: use storage_paths for root resolution, add proper
MIME types to FileResponse, add auth via query token for <audio> elements
- Fix assets_serving.py: add path traversal security check
- Fix step4_asset_routes.py: use get_user_workspace() instead of WORKSPACE_DIR,
detect actual audio format before saving preview
- Fix get_db() in database.py: raise HTTPException(401) instead of raw Exception,
catch engine creation failures with HTTPException(503)
- Fix avatar.py: add auth error handling, diagnostic logging for path resolution,
graceful DB save degradation
- Upgrade utils/storage_paths.py with robust find_repo_root() (env var override + validation + fallback)
- Remove broken _find_root() from podcast/constants.py, import from storage_paths instead
- Fix ROOT_DIR resolving to backend/ instead of project root (caused avatar upload 500s on Render.com)
- Fix video_combination_service.py default output dir (was writing to data/media instead of workspace)
- Add deprecation comments to global data/media constants in media_utils.py
- Pass user_id through resolve_media_path for tenant-scoped podcast resolution
- Add ALWRITY_ROOT_DIR env var support for explicit production overrides
- Log warning when get_podcast_media_dir called without user_id
- Use OperationButton with cost display for scene action buttons
- Voice clone integration: When user selects voice clone in Write phase,
backend uses their uploaded voice sample + scene script text to generate
audio via qwen3/minimax/cosyvoice voice clone APIs
- Multi-tenant workspace storage: All podcast assets (audio, video, images,
charts) now use workspace-specific directories per user
- Chart preview improvements: Card-based B-Roll charts UI with thumbnails,
takeaway text, and action buttons; public endpoint for image serving
- Voice clone caching: In-memory LRU cache for voice samples (avoids
re-downloading per scene); frontend caches voice clone metadata
- Thread pool for voice clone: Audio generation uses ThreadPoolExecutor to
avoid blocking the FastAPI event loop
- Auto-detect voice clone IDs (vc_*, MY_VOICE_CLONE) to route correctly
- DB fallback for voice sample URL: Fetches from ContentAsset if not passed
- Fixed API URL resolution for chart previews
- Fixed GlassyCard DOM warnings for motion props
- Fixed ScriptGenerationProgressView syntax error
- Fixed usePodcastWorkflow scriptData reference
