nemlai

CLI and Python SDK for automated grocery replenishment on Nemlig.com — Denmark's largest online grocery delivery service.

NemlAI predicts what groceries you'll need next based on your order history, then builds ready-to-checkout basket links. Especially useful for OpenClaw and other AI agents that interact with the NemlAI platform.

Platform: nemlai.milops.org

Install

pip install nemlai

On Arch Linux:

yay -S nemlai

On Ubuntu:

sudo add-apt-repository ppa:milops/nemlai
sudo apt install nemlai

Quick start

# Log in to your NemlAI instance
nemlai login

# Interactive mode
nemlai

# Run a single command
nemlai predict
nemlai search "mælk"
nemlai basket

Python SDK

from nemlai import NemlAI

client = NemlAI.login("you@example.com", "password")  # or NemlAI(token="...")

basket = client.basket.generate(size="medium")       # predicts, polls, returns JSON
for item in client.basket.items()["items"]:
    print(item["name"], item["quantity"], item["explanation"])
client.basket.adjust("5001234", 2)                    # by product_number
client.basket.compare()                               # price the basket at every enabled shop
client.basket.send()                                  # returns basket_url

Basket methods return JSON by default since 0.10.0 (remove/adjust take a product_number). Pass structured=False for the old text replies.

The basket at other shops

The same basket priced at Bilka, Netto, Føtex and REMA 1000 (the shop tabs on /kurv). Where a shop has no validated match you get a suggestion (status: "suggestion", with differs), and you can pick another product for this basket only:

prices = client.basket.shop_prices()                 # every shop tab, current basket
for shop in prices["shops"]:
    print(shop["name"], shop["total"], shop["counts"])

milk = client.basket.items()["items"][0]
alts = client.basket.alternatives("netto", canonical_id=milk["canonical_id"], quantity=milk["quantity"])
client.basket.choose(milk["product_number"], "netto", alts["items"][0]["product_number"])
client.basket.shop_prices(shops=["netto"])           # the pick comes back as status "chosen"

client.basket.send_to_shop("bilkatogo")              # your own BilkaToGo cart; never checks out

Picks are never stored on the server and are dropped by generate(). send_to_shop only works for shops whose send_mode is basket_push in client.basket.shops() (today BilkaToGo, after nemlai shops link).

nemlai basket prices                              # totals per shop
nemlai basket prices --shop netto                 # every line at Netto
nemlai basket alternatives 3 --shop netto         # or --search "rugbrød"
nemlai basket choose 3 --shop netto --product 123456
nemlai basket send-shop bilkatogo

Picks made in the CLI are kept in ~/.config/nemlai/basket_picks.json until the next basket generate.

Exclusions and other shops

Leave orders or single products out of predictions and analytics, and link Bilka/Føtex so their receipts count too:

nemlai exclusions list
nemlai exclusions items 4211
nemlai exclusions exclude 4211          # whole order
nemlai exclusions hide 4211 5001234     # one product in that order
nemlai shops status
nemlai shops link --email you@example.com --shops bilka,foetex   # prompts for the password
nemlai shops refresh bilka
nemlai settings update --cutoff 2026-01-01 --auto-learn true

Spending analytics

Ask where the grocery money went: from the CLI (nemlai analytics guide for the full reference) or from Python.

nemlai analytics overview --last this-month
nemlai analytics spend --product mælk --category Mejeri --last 12m --compare previous-period
nemlai analytics top categories --level sub --from 2026-01-01 --to 2026-03-31
nemlai analytics trend --bucket month --last 12m
nemlai analytics drill --order 0042
nemlai analytics receipts --product jordbær --last 3m --detail full --save
nemlai analytics categories-tree --last 12m

Tables are rendered by the SDK (canonical product variants, comparison changes, saved amounts, truncation notices); --format json|csv|markdown returns the server's own output.

from nemlai import NemlAI, AnalyticsError

client = NemlAI(token="your-token")
overview = client.analytics.overview(last="3m")
print(overview.spend.total, overview.spend.saved, overview.spend.comparison)

milk = client.analytics.spend(product="mælk", compare="previous-period")
for match in milk.matches:
    print(match.name, match.total, match.product_numbers)

try:
    client.analytics.trend("fortnight")
except AnalyticsError as err:
    print(err.status_code, err.message, err.hint)

Methods: overview, spend, top, trend, drill, receipts, categories_tree (typed dataclass results), plus text(endpoint, payload, format="csv") for server-rendered output.

What it does

• Fetches order history from Nemlig.com
• Uses a cloud ML model trained on your purchase patterns
• Predicts quantities for your next basket
• Builds Nemlig basket URLs you can review and checkout
• Supports manual item search and hybrid baskets (manual + ML)

For agents

NemlAI has an MCP server for AI agents (Claude, ChatGPT, Codex, Grok, Hermes, OpenClaw and others) at https://nemlai.milops.org/mcp, and a nemlai skill that teaches them the workflows. Sign in with OAuth where the client supports it, or with the token from nemlai login (stored in ~/.config/nemlai/token).

Connect in one minute

export NEMLAI_TOKEN=$(cat ~/.config/nemlai/token)   # the configs read the token from here
nemlai mcp install claude-code    # or: codex, hermes, openclaw
nemlai mcp doctor                 # checks the URL, the token (whoami) and the tool list

• nemlai mcp install <client> writes the client's MCP config and installs the skill. It backs up the file first, never changes other entries, and replaces an existing nemlai entry only with --force. --dry-run shows the diff; --embed-token writes the token itself instead of ${NEMLAI_TOKEN}.
• nemlai mcp config <client> prints the config instead (claude-code, codex, hermes, openclaw, claude-desktop, generic).
• claude.ai, Claude Desktop, ChatGPT and Grok: add a custom connector with the URL and sign in (OAuth, Giv adgang).

The server explains itself: its instructions and the nemlai_guide tool cover the data model, the rules (show the basket and total before sending, send only on the user's go-ahead, NemlAI never checks out, never put a password in chat) and the workflows.

When MCP fails, this CLI is the fallback. Every MCP tool has a CLI command, and --json prints one JSON object per command:

nemlai --json whoami
nemlai --json basket items
nemlai --json analytics spend --last 30d

If the CLI fails too, the web app at https://nemlai.milops.org has every feature. The full map (every tool's CLI command and web page), the rules and the workflows: https://nemlai.milops.org/agents.

License

MIT