tomas.kracmar
released this
2026-04-20 16:11:31 +00:00
| 48 commits to main since this release
What's New
AI Feature Flag (AI_FEATURES_ENABLED)
- Gate all AI/natural-language features with a single environment variable.
- When
false, the /api/ask endpoint is completely unregistered and the "Ask a question" panel is hidden from the UI.
- New
GET /api/config/features endpoint exposes feature flags to the frontend.
MCP Server (backend/mcp_server.py)
- Standalone Model Context Protocol server for Claude Desktop, Cursor, and other MCP clients.
- Connects directly to MongoDB (bypasses FastAPI auth layer — run only in trusted environments).
- Exposes four tools:
search_events — filter by entity, service, operation, result, and time range.get_event — retrieve raw event JSON by ID.get_summary — aggregated activity summary (by service, operation, result, top actors) for the last N days.ask — natural language query that returns recent matching events.
Natural Language Query (/api/ask)
- Ask questions like "What happened to device ABC123 in the last 3 days?"
- Intent-aware service filtering: broad queries automatically exclude high-volume Exchange/SharePoint noise.
- Smart sampling: when datasets are large, failures and high-value services are prioritized for LLM context.
- Aggregated overviews for datasets with >50 events.
- Respects active UI filters (services, actor, operation, result, tags).
- Azure OpenAI / MS Foundry compatible (
api-key header, api-version, max_completion_tokens).
Version Endpoint
GET /api/version returns the running version (baked into the Docker image at build time).- Displayed as a badge in the UI header.
Upgrade Notes
- Pull the new image:
docker compose pull && docker compose up -d
- Optionally set
AI_FEATURES_ENABLED=false in .env to disable AI features.
- Optionally configure
LLM_API_KEY, LLM_BASE_URL, LLM_MODEL for the /api/ask endpoint.
- For MCP: install
mcp from requirements.txt and configure your MCP client to run python backend/mcp_server.py.
Downloads