Queriva

Ask your data anything.

Self-hosted, LLM-powered semantic search. Ingest documents from any source, search in natural language, get AI-synthesized answers — zero external API calls.

Status: v0.4.0+ — full Docker Compose stack (qdrant, ollama, embed-sidecar, api, ui). See docs/SPEC.md for architecture and docs/ISSUES.md for the backlog (next: #28 E2E smoke → v0.5.0).

Quick start

docker compose up -d --build   # or: make compose-up
make ollama-pull               # first run only — downloads mistral (~4GB) for RAG mode
make seed                      # ingest fixtures/news_radar_dhaka_floods.json into news_radar

Then open http://localhost:3000 and search floods in Dhaka last week.

make seed validates the fixture, ingests 8 articles (4 Bangla + 4 English), and verifies idempotency. Requires API, embed-sidecar, and Qdrant from docker compose.

Dev compose (hot reload)

make compose-dev    # UI Vite HMR + API mvn spring-boot:run

Docs

• docs/SPEC.md — full specification
• docs/ISSUES.md — implementation backlog
• docs/adr/ — architecture decision records
• CHANGELOG.md — version history

Full README coming at v1.0.0 — see README-TEMPLATE.md.