OpenDataAnalysis — Data skill for Claude Code

Open Data Analysis

[中文说明](README.zh-CN.md)

Interactive data analysis for tabular data. Users upload CSV/Excel files or import SQL source snapshots, then an agent inspects data, runs SQL/Python when needed, creates charts, and produces reports.


Current Capabilities

• Agent runtime with tool calling and self-directed planning, not a fixed DAG.
• Auth, workspaces, sessions, runs, file ownership, and resumable reports.
• CSV/Excel upload plus PostgreSQL/MySQL workspace sources imported as bounded snapshots.
• Workspace semantic assets, general governance fact inspection, and audit events for critical data operations.
• Go backend, Vue 3 frontend, SQLite metadata DB, and per-session SQLite analysis DBs in development mode.
• Local object storage abstraction with explicit production guardrails for the S3-compatible migration.

Quick Start

Local development is standardized on Docker Compose.

cp server/.env.example server/.env
# Configure LLM_PROVIDER / LLM_API_KEY / LLM_MODEL / AUTH_SECRET and related settings.
docker compose up -d --build

Open `http://localhost`.

Common commands:

docker compose up -d --build --force-recreate
docker compose logs -f server
docker compose logs -f client
docker compose logs -f python-executor
docker compose down

Deployment Modes

The default Docker Compose setup is a development profile. It uses local SQLite metadata, local object storage, in-process run execution, session SQLite analysis scratch DBs, and executor-local Python artifacts.

Production must be configured explicitly with `DEPLOYMENT_MODE=production`. In that mode the server fails closed while any local or single-process backend is still selected:

• METADATA_STORE=sqlite
• STORAGE_PROVIDER=local
• RUN_BACKEND=inprocess
• ANALYSIS_STORE=session_sqlite
• PYTHON_ARTIFACT_STORE=executor_local
• wildcard or localhost CORS_ALLOWED_ORIGINS

The target MaaS contract is documente