Skip to content

Quickstart — the customer's data in minutes

Quickstart

Install lm, log in, load one export, show the rows, ask a question.

Before this page

The hub runs.

You need

One CSV export from the customer.

You get

Rows you can show, and a first answer in French or English (Pro).

Other sources

An API, an SFTP server, an S3 bucket: Other first sources.

1. Install lm

Take the archive for your platform from the Releases page. There are five: macOS (Intel, Apple Silicon), Linux (x86-64, arm64), Windows x86-64.

shasum -a 256 -c lm_<version>_checksums.txt --ignore-missing   # check the download first
tar -xzf lm_<version>_darwin_arm64.tar.gz                      # your platform's archive
install -d ~/.local/bin && install -m 0755 lm ~/.local/bin/lm
  • Windows: the archive is a .zip with lm.exe. Unpack it into a folder on your PATH.
  • From source (Go 1.26+): cd lm && make install.
  • The shell cannot find lm? Add ~/.local/bin to your PATH.

Check it:

$ lm version
lm version dev (commit 94ae18cc9667, built from 2026-10-19T05:34:15Z)

A downloaded binary prints its release number instead of dev. Put this line in a bug report.

2. Log in

lm config set-context selfhost --server http://localhost:8080 \
    --oidc-issuer http://localhost:8180/realms/lumnik
lm config use-context selfhost
lm login

lm login prints a link and a code, then waits. Open the link in a browser. Sign in as integrator: the password is INTEGRATOR_PASSWORD in .env. The terminal then finishes:

$ lm login
To sign in, open:
  http://localhost:8180/realms/lumnik/device?user_code=EYQP-LLJN
and enter code: EYQP-LLJN

Waiting...
Logged in to lumnik Open Edition.

On a Pro hub, the last line reads Logged in.

Check the round trip. On a fresh hub, the table is empty:

$ lm connector list
NAME  TYPE  TENANT  SCHED  WINDOW  STATUS  LAST RUN

3. Load the export

The customer hands you stock.csv:

sku,label,qty,warehouse
PMP-100,Pompe doseuse,12,Lyon
VAN-200,Vanne papillon,0,Lyon
FLT-300,Filtre a cartouche,40,Grenoble

One command loads it, in a scope you name:

$ lm csv ingest stock.csv --scope retail --name stock-count
Ingested stock.csv into ext.stock_count
  • rows written:    3
  • promoted cols:   4
      - sku (text)
      - label (text)
      - qty (bigint)
      - warehouse (text)

  • askable:         scope 'retail' — try: lm ask --scope retail "how many rows are in ext.stock_count?"
A dormant connector is registered — see `lm connector list`.
  • qty holds numbers only, so it lands as a number. The other columns land as text.

  • On lumnik Open, the last two lines read:

    A one-shot: no run recorded, nothing indexed for chat.
The rows are in ext.stock_count, and a dormant connector is registered — see `lm connector list`.

  • The load runs once. Nothing runs again on its own.

4. Ask a question (Pro)

Ask in the scope you named:

$ lm ask --scope retail "quels articles ont une quantité à zéro ?"
Les articles ayant une quantité à zéro sont :

- SKU : VAN-200
  Libellé : Vanne papillon

  SKU      LABEL
  VAN-200  Vanne papillon
  1 row returned

  SQL: SELECT "sku", "label"
FROM "ext"."stock_count"
WHERE "qty" = 0
LIMIT 500
  • The rows and the SQL always come with the answer. Show them to the customer.
  • Use the words of the data. « Quels articles sont en rupture de stock ? » is declined: no column says "rupture". « Quantité à zéro » names the column, and works.
  • Without --scope, the question goes to your default scope, where this file is not.
  • On lumnik Open, lm ask answers Error: this hub's build has no Ask: …. On Pro with the AI switched off, it answers Error: this hub has the AI switched off ….

Why Ask declines, and what to do: When Ask refuses.

5. Take it back off

$ lm connector delete stock_count --dry-run
Would delete connector stock_count (csv-file) — 3 rows of ext.stock_count, its run history and RAG chunks (dry run, nothing deleted)

Without --dry-run, lm asks you to type delete. Any other answer deletes nothing. A script passes --force.

Next

  • Make it recurring: a csv-file connector reads the export on a schedule: CSV & files. Give its manifest the name lm connector list shows: --name stock-count was registered as stock_count. Then lm connector resume stock_count.
  • Another source: Other first sources — an API, SFTP, S3.
  • The full demo, with two sources fused into one customer: The WOW.

What makes a scope answerable

Ask needs to know a scope's tables. A connector run, a one-shot load or an entity apply tells it, with nothing to configure. See what Ask knows of a scope:

lm source schema --scope retail

Empty, or not found? Rebuild it with lm source rediscover --scope retail, then ask again.

Troubleshooting

You see Do this
Error: API error 401 … on every command Your session ended. Run lm login again.
Error: access refused (403): scope not granted Your account holds scope: roles, and not this one. Use a scope you hold.
Error: this hub's build has no Ask … or Error: this hub has the AI switched off … on lm ask lumnik Open, or the AI is off. See The two editions.
Ask declines a question Rephrase it with the words of your columns. Check lm source schema --scope <tag>.
A second connector appears after lm apply -f metadata.name differs from the registered name. Delete the duplicate, re-apply under the name lm connector list shows.