Loom is in pilot intake · these docs cover the full data-science lifecycle CLI Talk to us
Docs/Get started/Getting started

Getting started

Loom is a small, polyglot install — the loom command (Node), the Python engine it drives, and a local datastore for the lifecycle verbs. One command sets all three up; then you just talk to it.

Install (macOS)

Install the CLI globally with npm, then run it:

npm install -g @zkailabs.com/loom
loom

The npm package is a tiny launcher. On its first run, loom clones the Loom repo with your own git credentials, runs the installer (the engine, the CLI build, and a local Metaflow datastore), and from then on delegates every command to the installed CLI.

i

Loom is currently private — installing it requires git access to the repository (an SSH key authorized for the org). That access is the gate; without it the first run stops with a clear message. Need access? Reach out.

Prefer to do it by hand?

If you've already cloned the repo, run the installer directly — it prefers an AI assistant (Claude or Codex) to drive the steps and adapt to your machine, and falls back to a plain script:

# clone, then:
./install.sh

Run it

Open the agent and ask for what you want in plain English — Loom picks the right verb, runs it, and composes the next step:

loom                          # open the agent (interactive)
loom "profile my data and flag leakage"
loom eda --dataset IngestDataset/123   # or jump straight to a verb

Inside the agent, every verb also has a /loom-<verb> slash-command, and you can toggle plan mode with /plan. See the agentic CLI for the full picture.

Stand up the datastore

The lifecycle verbs read and write through a local Metaflow datastore (minikube + minio, set up by the installer). Start its port-forward (keep it running) and source the environment:

kubectl port-forward -n loom svc/minio 9000:9000 9001:9001 &
source .env.metaflow

You can browse what's stored at the minio console (http://localhost:9001); these are local-dev credentials baked into the manifest, never anything that leaves your machine.

Verify with loom doctor

Health-check the whole stack — Python engine, Metaflow, datastore reachability — read-only:

loom doctor          # expect: VERDICT: PASS

If a check fails, loom doctor --fix hands the failures to Claude or Codex to resolve (following the install guide), falling back to manual hints when neither is installed.

i

The read-only and lifecycle verbs (eda, features, validate, viz, report, …) work without a model key. Only the natural-language planning and the search engine need one — a missing key gives an actionable line, never a crash.

Keeping current

Update the whole stack — pull the latest, rebuild the CLI, refresh the engine — with one command:

loom update