Loom is in pilot intake · these docs cover the full data-science lifecycle CLI Talk to us

Loom documentation

Loom is an agentic CLI for the whole data-science lifecycle — talk to it in plain English and it picks the right verb, runs it, reads the structured result, and composes the next step. Start here, then dive into any feature in detail.

Start here

Getting started

Install on macOS in one command, run your first verb, and verify the stack with loom doctor.

Drive Loom

The agentic CLI

How loom turns plain English into the right verb — the REPL, slash-commands, streaming, and approval gates.

Drive Loom

Lifecycle & verbs

The full verb catalog — connect, eda, features, optimize, validate, train, deploy, ops — and the approval tiers.

Drive Loom

Plan mode

A read-only exploration phase: explore the data, propose a lifecycle plan, then approve and execute.

Drive Loom

Subagents

Delegate to focused agents — data-scout, pipeline-builder, result-reviewer — each tier-safe by design.

Drive Loom

Data access (MCP)

Reach data across vendors and clouds, and drive cloud ops, through the Model Context Protocol.

Drive Loom

The workbench

Web search & PDFs, document parsing, inline charts, markdown/LaTeX, diagrams, session search, memory.

Concepts

The data model

Your input is a Metaflow data object, referenced by pathspec. One ingest boundary; data stays in your perimeter.

Reference

Configuration & providers

Every flag and environment variable, the swappable provider ports, and the model-provider options.