Open Eyes

GenAI-powered open government data dashboard. Pre-ingests German open government data (GovData.de, ~149K datasets) into DuckDB, then lets users ask natural language questions via a chat sidebar. An LLM generates SQL queries, executes them, and renders interactive ECharts visualizations — making open data truly accessible.

Built in Rust/Dioxus, starting with Germany, extensible to other EU CKAN portals.

Architecture

┌──────────────────────────────────────────────────────────┐
│                    User Browser (WASM)                     │
│  ┌─────────┐  ┌──────────────────┐  ┌─────────────────┐  │
│  │ Sidebar  │  │   Main Content   │  │   Chat Panel    │  │
│  │  Nav     │  │  (Home/Catalog/  │  │  NL Question →  │  │
│  │         │  │   Analytics)     │  │  Charts + Text  │  │
│  └─────────┘  └──────────────────┘  └─────────────────┘  │
└──────────────────────┬───────────────────────────────────┘
                       │ Server Functions
┌──────────────────────▼───────────────────────────────────┐
│                   Axum Server (SSR)                        │
│  ┌─────────────┐  ┌─────────────┐  ┌──────────────────┐  │
│  │  DuckDB     │  │  LLM Client │  │  CKAN Client     │  │
│  │  (embedded) │  │  (OpenAI    │  │  (pre-ingestion) │  │
│  │             │  │   compat.)  │  │                  │  │
│  └─────────────┘  └─────────────┘  └──────────────────┘  │
└───────────────────────────────────────────────────────────┘

Workspace Crates

Crate	Purpose
open-eyes-core	Config, models, DuckDB pool, LLM client, EChart option builders
open-eyes-ingest	CLI for CKAN crawling + resource download + DuckDB loading
open-eyes-dashboard	Dioxus fullstack app (web + server features)

Quick Start

Prerequisites

• Rust 1.84+
• Dioxus CLI (cargo install dioxus-cli@0.7.3)

1. Configure

Copy and edit the config file:

cp config.toml config.local.toml
# Edit config.local.toml with your LLM API key

Or use environment variables:

export OPEN_EYES_LLM_API_KEY="your-api-key"
export OPEN_EYES_LLM_BASE_URL="https://api.openai.com/v1"  # or any OpenAI-compatible endpoint
export OPEN_EYES_LLM_MODEL="gpt-4o-mini"

2. Ingest Data

# Crawl 10 datasets from GovData.de
cargo run -p open-eyes-ingest -- crawl --max-datasets 10

# Download and load CSV/JSON resources into DuckDB
cargo run -p open-eyes-ingest -- load

3. Run Dashboard

dx serve --package open-eyes-dashboard --platform web

Open http://localhost:8080 in your browser.

4. Ask Questions

Click the chat button (bottom-right) and ask questions like:

• "Welche Datensätze hat das Statistische Bundesamt?"
• "Show me the top 10 categories by dataset count"
• "What are the most common data formats?"

Docker

# Build and start the dashboard
docker compose build
docker compose up dashboard

# Ingest data (one-off)
docker compose --profile ingest run ingest-crawl
docker compose --profile ingest run ingest-load

Environment Variables

Variable	Default	Description
OPEN_EYES_LLM_API_KEY	(empty)	LLM API key
OPEN_EYES_LLM_BASE_URL	https://api.openai.com/v1	OpenAI-compatible endpoint
OPEN_EYES_LLM_MODEL	gpt-4o-mini	Model to use
OPEN_EYES_DUCKDB_PATH	./data/open-eyes.duckdb	Database file path
OPEN_EYES_PORT	8080	Server port
OPEN_EYES_CKAN_BASE_URL	https://ckan.govdata.de/api/3/action	CKAN API endpoint

Development

# Check all crates
cargo build -p open-eyes-core
cargo build -p open-eyes-ingest
cargo build -p open-eyes-dashboard --features server

# Run tests
cargo test -p open-eyes-core
cargo test -p open-eyes-ingest

# Lint
cargo fmt --all
cargo clippy -p open-eyes-core -p open-eyes-ingest -- -D warnings
cargo clippy -p open-eyes-dashboard --features server -- -D warnings

License

MIT