Better Claude Code for data engineers.
Everything you love about Claude Code, skills, MCP, self-evolving memory, rebuilt for the data stack. Model-neutral, governed, and wired into your warehouse from the first command.
All of Claude Code's power, in your data terminal.
Datus CLI keeps the general agent capabilities you already rely on, and adds nothing you have to relearn.
Self-evolving agent
Skills, knowledge, and memory that iterate on their own, every run sharpens the next.
MCP & Skills, in full
Complete support for the MCP ecosystem and the Skill system. Bring any tool, ship any skill.
IM Gateway
Drive the agent from where your team already works, Slack and Lark channel integration built in.
Model-neutral by design.
Not locked to one vendor. Route each subagent to the model that fits its job, or run an all-Claude lineup. See the full supported-models list in the docs.
Supported modelsPrefer one family? Mix Claude Haiku, Sonnet, Opus, and Fable across the same agents, your call.
Guardrails the data stack actually needs.
Agents touch production data. Datus puts three layers of real control between the model and your warehouse.
Tool permission modes
How much power the agent gets, set per session or per subagent:
normalRead-onlyExplore, query, and plan. Nothing gets written.autoEditingEdits and ships within its lane, no prompt per step.dangerousData sandboxFull-power operations, contained to a sandbox.Query-time scoping
What a subagent can see at run time, scope it down to just:
Data access policy
Enforced outside the model, on every query the agent runs:
Built for the big-data ecosystem.
Warehouses, semantic layers, schedulers, BI, Datus plugs into the stack you already run, and its built-in skills and subagents cover the whole development lifecycle.
Connect your stack
Snowflake, Databricks, StarRocks, ClickHouse, MetricFlow, Airflow, Superset, Grafana, and more.
Full-lifecycle skills
Built-in skills and subagents take you from exploration to metrics to production ETL, end to end.
One evolving context
Schemas, semantics, and validated SQL captured into portable memory that powers every integration.
Frequently asked questions
Install, models, Subagents, and how the CLI relates to Studio.
What do I need to install Datus CLI?
You need Python 3.12 or newer and pip. Install with pip install datus-agent, then run datus-agent to start the interactive CLI. Configure your LLM API key and database connection in agent.yml. Datus CLI works on macOS and Linux today; Windows is supported via Python but not officially certified. Full setup steps are in docs.datus.ai Quickstart.
Does Datus CLI work with my existing warehouse?
Yes. Datus ships native adapters for Snowflake, PostgreSQL, MySQL, ClickZetta, and others, plus MCP-based connectors for DuckDB, StarRocks, Hive, Spark, ClickHouse, and Trino. You point the CLI at your catalog or JDBC connection—no need to migrate data. Custom DB adapters can be added via the plugin architecture described in the GitHub repo.
How is Datus CLI different from Datus Studio?
Datus CLI is the full open-source agent for engineers: context building, Subagent creation, MCP tools, and local control. Datus Studio is the hosted web experience for faster trial and chat-style exploration. Both share the same Context Engine concepts; many teams prototype in Studio and run production workflows in CLI or Enterprise.
Can I use my own LLM API keys with the CLI?
Yes. Datus CLI is bring-your-own-key. You configure OpenAI, Claude, Qwen, DeepSeek, Kimi, Gemini, or others in agent.yml, including per-Subagent model overrides. Usage and cost stay on your provider account. Cloud Personal may offer managed keys; the open-source CLI never requires a Datus-hosted model.
What can I build with Subagents in the CLI?
A Subagent is a scoped chatbot backed by roughly ten tables, twenty metrics, and thirty reference SQL patterns for one business domain. You create them with .subagent add, refine context through feedback loops, and export mature Subagents as HTTP APIs or MCP servers for other agents to call.
Fine, we know you still love Claude Code, you can use datus-mcp right inside it. Learn how ↗
Start with the open-source CLI.
Free, Apache-2.0, your data and model. Five minutes from install to your first validated query.
Bring your own warehouse & model