Datus-ai/Datus-agent

Datus · Open‑Source Data‑Engineering Agent

What it is – Datus is a Python‑based AI‑augmented assistant that helps teams work with the modern data stack. It connects to your data warehouse, catalog, semantic layer and BI tools, and uses large language models (LLMs) to generate and validate SQL, build semantic models, create ETL pipelines, and produce reports or dashboards. Every interaction is stored in a context engine so the agent gets more accurate over time.

How it works – The system is split into three user‑facing entry points:

• CLI (datus) – an interactive REPL for data engineers.
• Web Chat (datus --web) – a FastAPI + React chatbot accessible from a browser, Slack, Feishu, or VS Code.
• REST / MCP API (datus-api, datus-mcp) – programmatic access for other applications.

Under the hood, a core composed of subagents (domain‑specific knowledge packages) and skills (plug‑in tools) talks to a context engine that stores schemas, reference SQL, business rules and other artefacts in an embedded LanceDB/SQLite (or PostgreSQL for teams). Adapters let the core talk to 21+ databases, 10+ LLM providers, job schedulers (Airflow), and BI platforms (Superset, Grafana). A separate semantic engine called Dosi compiles a unified semantic model to SQL for more than a dozen dialects.

Key features

• Automated semantic modeling – reads your schema & query history, then creates OSI‑compatible semantic models and metric definitions without hand‑written YAML.
• Metric Q&A – answers business questions using the metric layer, providing attribution when a metric changes.
• Self‑improving context – corrections are written back to the knowledge base, improving future responses.
• Built‑in subagents for cross‑database migration, ETL job generation, wide‑table builds, and dashboard/report creation.
• Open ecosystem – plug‑in framework (datus-plugin.yml) lets you add new CLI commands, skills, or external integrations; a marketplace follows the agentskills.io convention.
• Enterprise governance – role‑based permissions, SQL pre‑review, OS‑level sandboxing, and OTLP‑compatible tracing.

Getting started (Linux/macOS, Python 3.12+)

# one‑liner installer
curl -fsSL https://raw.githubusercontent.com/datus-ai/datus-agent/main/install.sh | sh

# start the REPL
datus
# then inside the REPL
/model          # pick an LLM
/datasource     # add a warehouse connection
/init           # optional: scan the current project

Alternatively install from PyPI: pip install datus-agent (or faster with uv). The docs include a full tutorial that builds a pipeline on a sample “California schools” dataset.

Typical use cases

• Data engineers exploring a new warehouse, building reusable semantic models, and generating ETL jobs.
• Business analysts asking metric‑level questions via chat, receiving SQL‑free answers and visual dashboards.
• Application developers consuming data‑service endpoints through the REST or MCP API.
• Teams extending the platform with custom skills or adapters without modifying the core.

License – Apache 2.0.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.