Datus, a data engineering agent that builds evolvable context for data system (github.com)

Datus-agent is an open-source data-engineering agent that brings a Claude-Code-like command-line experience to data teams: Datus-CLI for interactive SQL, agentic actions and context building; Datus-Chat for a multi-turn web chatbot with upvotes and issue reporting; and Datus-API to expose stable data services. Rather than just building pipelines and tables, Datus automatically constructs a living semantic map of your company’s data by combining metadata, metrics, SQL history and external knowledge. Engineers can import SQL history, generate semantic models (/gen_semantic_model), iterate in the terminal (/chat, @table, @file), and spin up scoped subagents that encapsulate the right context, tools and rules for a specific domain. Installation requires Python >= 3.12 (pip install datus-agent==0.2.1; datus-agent init). This matters because it shifts collaboration from raw SQL handoffs to reusable, domain-aware assistants that learn from real feedback: analysts upvote answers, file issues or export results, and engineers refine SQL and rules so each interaction improves reasoning accuracy. Subagents can be limited in scope (e.g., a handful of tables) and deployed to a web UI for analysts, turning data domains into self-evolving chatbots. For AI/ML teams this lowers friction for data consumers, embeds governance and context into queries, and creates an auditable feedback loop that can improve correctness and reuse over time.