- Python 100%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01X3rSFQDJYjX6bGw3aeA4iy |
||
| data/restaurants | ||
| src/noms | ||
| tests | ||
| .gitignore | ||
| LICENSE | ||
| pyproject.toml | ||
| README.md | ||
| uv.lock | ||
noms
CLI menu roulette + dairy/calorie tracker over cached, refreshable restaurant data.
Pick a place, pick a dish, or let it roll for you — filtered to what you can actually eat (dairy-aware, with provenance), and log where you went even when you don't remember what you ate.
Why the schema looks like a POS system
"Regular vs Small with different calories" and "safe if you ask for no cheese" are the same problem every restaurant point-of-sale solves. An item has:
- variants — sizes/portions, each with its own nutrition (Regular / Small)
- modifier groups — choices that shift calories and dietary flags (protein swap, "no cheese", "add guac")
So the dairy verdict can be computed ("safe IF you skip the cheese modifier") and carries
provenance: OFFICIAL_ALLERGEN > OFFICIAL_MENU > STAFF_CONFIRMED > ASSUMED >
UNCONFIRMED. refresh re-pulls calories/prices from the source URL but never clobbers a
human-verified dairy verdict.
DairyStatus: SAFE_AS_IS · SAFE_WITH_MOD · CONTAINS · UNKNOWN
Install / run
cd noms
uv sync # core CLI
uv sync --extra api # + FastAPI service
uv run noms list # what's cached, freshness, % dairy-edible
uv run noms show noodles_and_co --dairy safe
uv run noms roll # roll restaurant + dish (dairy-safe default)
uv run noms roll chipotle --max-cal 700 # roll within one place, calorie cap
uv run noms roll --tag vegetarian --n 3 --log # 3 picks, log them
uv run noms roll cava # → "👉 hold the tzatziki, feta"
uv run noms guess "rice, beans, chicken, cheese, sour cream — 600 cal" # classify pasted text
uv run noms log "McAlister's" -n "team lunch, didn't track item" # item-less visit
uv run noms log mcalisters --item "Sweet Tea" # auto-fills cal/dairy from cache
uv run noms history
uv run noms stats
uv run noms refresh noodles_and_co # re-fetch + merge (preserves dairy work)
uv run noms serve # FastAPI on :8723
Data
data/restaurants/<slug>.yaml — one file per restaurant, editable by hand. Override the
location with NOMS_DATA_DIR. The eat log is append-only JSONL at <data_dir>/log.jsonl.
Seeded with Noodles & Company, LoLo's Bowls, McAlister's Deli, Chipotle, CAVA, and sweetgreen — chosen to exercise every schema corner (multi-size variants, protein modifier groups, all four dairy states, default-on dairy modifiers that drive the "hold the ___" advice, and mixed confidence). Calories are chain-published-approximate; dairy verdicts cross-checked against allergen guides where available.
noms guess — heuristic classifier
Paste an ingredient list / description and get a dairy verdict + any calories it can scrape
from the text. Same engine refresh uses; always ASSUMED confidence, so verify before
trusting:
noms guess "spiralized zucchini, tamarind sauce, peanuts, 410 calories" --name "zoodle thai"
# zoodle thai ✓ safe (assumed) — no dairy keywords found · calories detected: regular 410
"hold the ___" advice
When a SAFE_WITH_MOD item carries modifier options flagged sets_dairy: CONTAINS, default: true, roll/show compute the removal list instead of relying on a hand-typed
note — e.g. CAVA's Greek Bowl → 👉 hold the tzatziki, feta. SAFE_AS_IS items with
optional (default-off) dairy add-ons get the inverse: safe as-is — just skip optional ___.
API
GET /restaurants · GET /restaurants/{slug} · GET /roll?dairy=safe&max_cal=700 ·
POST /log · GET /log · GET /stats
Stack
uv · Typer (Click under the hood, by the FastAPI author) · Rich (dense tables) · Pydantic v2 (the schema) · httpx + selectolax (scrape) · FastAPI (optional service).