Local-first tool for profiling, exploring, and querying local data files (CSV, TSV, Parquet, JSON, JSON Lines, NDJSON, and DuckDB) with fast browse-and-import for .duckdb files via Import DuckDB (one system file dialog when native pick is enabled; HTTP upload remains for API/tests/fallback). Built for a single trusted workstation—fast EDA and ad-hoc DuckDB SQL, not hosted BI or multi-tenant use. This app is local-only; see SECURITY.md for the threat model and vulnerability reporting.

Quick start (no LLM required)

1. From the repo root: make install then make dev (requires GNU bash; see Platform notes).
2. Open http://127.0.0.1:5173.
3. New here? Follow the 5-minute tour with examples/ fixtures before using your own data.

Explore Columns, Charts, SQL, and more. Ask is optional and needs Ollama; see User guide — Ask.

Platform notes

• macOS — primary platform; Node 22 (.nvmrc), Python 3.11+, uv, optional Ollama.
• Linux — same make targets; install Node 22 and uv from your distro or upstream.
• Windows — use WSL2 (e.g. Ubuntu). Native Windows without WSL is untested.

Prerequisites: Node 22 from .nvmrc (matches CI). Python 3.11+ and uv. Run make help from the repo root for all targets.

Single-server mode

Opens http://127.0.0.1:8000 (API serves the built UI via DCC_UI_DIST_PATH). Day-to-day development uses make dev (Vite on 5173 + API on 8000).

Upgrading / workspace schema

Workspace metadata lives in DCC_WORKSPACE_DB_PATH (default .dcc_workspace.duckdb relative to the backend cwd). There is no in-place migration. After pulling changes that alter workspace layout or profile shape, run make clean-local or delete the workspace file by hand—that removes app cache, Ask history, and upload copies under .dcc_uploads/, not your original source files. See CHANGELOG for breaking changes. Schema details: backend/README.md.

Upgrading to 1.1.0

If you used 1.0.0, read CHANGELOG — 1.1.0 before upgrading. In short:

1. Update custom API clients: DuckDB inspect/import APIs use source_id (replaces upload_id).
2. /api/saved-charts was removed; saved chart rows are dropped from local workspaces on next startup.
3. Saved SQL snippets gain an optional description column; existing workspaces migrate in place on startup.
4. Large datasets may rebuild profiles with sample-scoped metrics; the UI polls prepare jobs until ready.

Upgrading to 1.0.0

If you used 0.1.0, read CHANGELOG — 1.0.0 before upgrading. In short:

1. make clean-local (or delete .dcc_workspace.duckdb) so the workspace and profile cache match the new schema.
2. Re-open datasets; profiling is async on first load (PROFILE_NOT_READY + job polling).
3. Custom Ask clients must call POST /api/agent/ask/stream only (sync ask removed).

1.0.0 also changes default UX: Ask runs a second Ollama summarization pass by default (set DCC_AGENT_SUMMARIZE_WITH_LLM=false for lower latency). Columns opens sorted by column name ascending; distribution stats show STDEV for standard deviation.

API reference

With the backend running: http://127.0.0.1:8000/docs (Swagger UI).

Architecture

• Frontend (frontend/): React, Vite, TypeScript, TanStack Query/Table, Zustand, ECharts, Tailwind, shadcn-style primitives
• Backend (backend/): FastAPI, DuckDB, Polars, Pydantic, uv

Documentation map

Known limitations (MVP)

• Excel and remote files are not supported yet.
• No cross-dataset join UI; explore overlaps with ad-hoc SQL.
• Very wide files may be slow on first profile; use Refresh in the dataset strip or POST /api/datasets/{id}/profile/refresh.