PIKA
Changelog
Release notes for PIKA — the private document intelligence system.
0.3.0 - 2026-02-09
Added
- Version display in health endpoint and admin About section
- Version detection tests for health endpoint and admin UI
- Comprehensive README sections for testing, CI/CD, and deployment
Changed
- Production hardening improvements for version management
- CI pipeline now runs on all
v*tags
Fixed
- Auth mock in
test_admin_page_accessible_when_authenticated
0.2.0 - 2026-02-09
Added
- Authentication & multi-user support — bcrypt password hashing, role-based access (admin/user), per-user chat history and sessions
- Query queue system — FIFO queue with configurable concurrency, per-user limits, position display, and automatic timeout
- Backup & restore — full ZIP backup of documents, index, config, and user database with background progress tracking and one-click restore
- Streaming responses — real-time token streaming from Ollama to the browser
- Prometheus metrics —
/metricsendpoint with request latency, query counters, queue gauges, and index stats - Structured JSON logging — production-grade logging with human-readable fallback in debug mode
- Ollama retry logic — automatic retry with exponential backoff for transient LLM failures
- Async background indexing — start/cancel/poll indexing operations without blocking the UI
- GitHub Actions CI — lint (ruff), test (pytest + coverage), dependency audit (pip-audit), and Docker image build/push to GHCR
- SSH deploy workflow — manual-trigger deployment to multiple environments with health checks and auto-rollback
- Release workflow — auto-create GitHub Releases from
v*tags - CSRF protection on all forms
- Rate limiting on queries and uploads via slowapi
- Security headers on all responses (X-Content-Type-Options, X-Frame-Options, X-XSS-Protection, Referrer-Policy)
- Audit logging for admin actions with pagination
- Feedback system — thumbs up/down on query responses
- Dark mode and modern UI redesign
- First-launch setup flow and admin page redesign
- NVIDIA GPU support in Docker Compose with AMD ROCm as commented alternative
- Automated backup via API with cron integration
- Graceful degradation when Ollama is unavailable
- Clear chat history button
- Cancel model download capability
- Loading indicators on admin documents section
- Admin page performance caching
Changed
- Default model changed to
llama3.2:3b - Consolidated Docker Compose files (base + override pattern)
- Hardcoded values moved to
config.py - Query status tracking is now user-specific
Fixed
- ChromaDB readonly database error after backup restore
- Streaming query source serialization (
asdict()) - History logging and navigation-resilient queries
- Race condition on
_running_queriesset - Restore success message now correctly states index is restored
- Duplicate message bug in chat
- Model pull progress tracking across page navigation
Security
- Password complexity requirements (min 8 chars, uppercase, lowercase, digit)
- Path traversal prevention on file operations
- Container runs as non-root
pikauser - Vulnerable dependencies bumped (pip-audit)
0.1.0 - 2026-01-15
Added
- Initial release of PIKA (Private Intelligent Knowledge Assistant)
- FastAPI web application with Ollama LLM integration
- RAG engine with ChromaDB vector storage and sentence-transformers embeddings
- Document processor supporting PDF, DOCX, TXT, and Markdown
- Web chat interface with file upload
- Docker deployment with setup script
- Admin page for model and document management
- Confidence scoring on query responses