AI Activity Dashboard Expansion & Privacy Enhancements

We've implemented phase two of the AI Activity Dashboard with a comprehensive set of analytical tools. Users now have access to new widgets, including:

• Focus on Activities: Work vs. Communications time distribution with change dynamics.
• Focus on Tasks: Distribution between Features and Bugs with percentages.
• Focus on Projects: Effort distribution across projects.
• Focus/Logged Time Metric: Ratio of focused to logged time.
• View as a List: Detailed list of all activities with sorting, filtering, and clickable links to tasks/MRs/projects.

We've also introduced an adaptive activity weighting system for different roles and accurate Productivity and Engagement calculations for managers:

• Engineers: Focus on Code, Tasks, Projects.
• Managers (Stakeholder, DM, PM, CTO, HR, Other): Adapted metrics for calculating time spent on documentation, meetings, and planning.

The dashboard includes the following technical improvements:

• Added Markdown support in AI Summary.
• Implemented user timezone support (instead of UTC 0).
• Unified tooltips in Week/Month view.
• Fixed multiple bugs (403 error for roles, table order, project IDs instead of names, diagram rounding, visibility of all projects, Activity logs sorting).
• Fixed a race condition when quickly switching months before View List.
• Implemented unified protocol for activity transfer to the activity service.

Impact: Managers and employees across all roles now have a fully enhanced tool for deep work focus analysis. You can now see not only overall activity but also detailed time distribution between work types (development vs. communications for engineers, meetings vs. documentation for managers) across projects and task types. Additionally, adapted metrics for different roles ensure objective productivity assessment regardless of position. The detailed View List allows drill-down into any activity for investigating anomalies or preparing reports, creating a comprehensive system for monitoring team performance and well-being that accounts for each role's work specifics.

Integrated personal data sanitizer for privacy protection. We've implemented personal data sanitizer integration into the data processing pipeline before sending to LLM. The system now automatically masks personal information (names, emails, phone numbers, addresses, and other PII) in texts sent to AI models for processing, with a toggle on/off capability. The implementation has been tested across various PM Agent and periodic task use cases.
Impact: Customers now have compliance with privacy requirements and GDPR; when sending context from chats, tasks, commits, and meetings to LLM for generating summaries and reports, employee and client personal data is automatically masked. This is critical for enterprise clients in regulated industries (banking, healthcare, and the government sector) where PII leakage to third-party LLM services is unacceptable. The sanitizer ensures a balance between AI functionality and compliance requirements, allowing safe use of powerful models without privacy violation risks.

Fixed project context selection in the PM Agent chat agent. We've eliminated the context determination error. The chat agent now correctly adds context from the project specified by the user or selected in the interface instead of mistakenly using data from another project. We've also fixed routing logic and context extraction in the request processing pipeline.

Fixed table display in HTML email reports. We've resolved a critical table rendering issue in email notifications; tables now display correctly in Markdown in PM Agent chat and properly convert to HTML in periodic tasks report emails. We've also fixed the logic for converting Markdown tables to HTML tables while preserving formatting and data structure.

Added API key verification and diagnostics system for Copilot. We've implemented a system for diagnosing API key status, providing information about the key name being used, key validity, key balance, and debugging information about system state. The PM Agent now explicitly reports key issues in chat (for example, "key ran out of funds" or "key is invalid"), allowing users to independently identify problems.

Optimized Copilot logging with environment variable control. We've revised the logging strategy in Copilot and added the LOG_LEVEL environment variable with levels WARNING (default for prod, only ERROR and WARNING), INFO (ERROR, WARNING, INFO), and DEBUG (all logs). For the production environment, we've removed excessive info and debug logs, keeping only critical ERROR and WARNING for problem monitoring.

Added Telegram Topics support for targeted alert delivery. We've implemented basic Telegram Topics support in the standup bot. The bot now accepts standups and reacts in any chat subgroup (topic). Alerts about needing to submit standups are now sent to specific topics (using the topic ID already received by the bot) instead of cluttering the General topic or entire chat. This ensures targeted communication in structured Telegram chats with topics.
Impact: Teams with organized Telegram chats now have clean communication: standup alerts arrive in a dedicated topic (e.g., Test/StandUp) without interfering with discussions in General or other topics. Additionally, the bot works correctly with subgroups, accepting standups from any topic and responding there, which is critical for large teams using Telegram Topics for communication segmentation.

Restored carryover vacation display when transitioning to a new year in Absence Overview. We've fixed a critical automatic vacation rollover error: when selecting 2026 in the Absence Overview filter, each employee's actual unused vacation days (carryover) from the end of 2025 now display correctly. Data automatically loads for all employees with remaining days; values correspond to actual unused days. HR managers now have accurate data for planning vacations in the new year; we've eliminated the critical problem of losing carryover information when changing years.

Fixed Labor Costs report generation for PMs without finance access. We've resolved the issue of generating empty Labor Costs reports for Project Managers when financial data viewing is restricted. PMs now receive reports with non-financial metrics while financial indicators are hidden, instead of empty reports. We've also ensured correct data filtering by access rights while preserving useful information.