Quickstart

costQL turns a live GraphQL API into a pricing pack (one self-contained JSON file) and then quotes any query against that pack fully offline. Build once in Python; quote forever in Python or JavaScript. This page gets you from zero to your first quote in about a minute, and to your first own pack in about fifteen.

1. Quote a committed demo pack (offline, 60 seconds)

pip install costql
git clone https://github.com/shapemachine/costql && cd costql

costql quote --pack packs/tmdb_t3.json '{ movie(id:"27205"){ cast(limit:8){ person{ name } } } }'

pricing pack: schema 26c786209ec27586 · tier T3 · currency work_ms · offline

  query      : { movie(id:"27205"){ cast(limit:8){ person{ name } } } }
  tier/basis : T3 · predicted
  price      : 50.5 work_ms   (safe max, never under real cost)
  typical    : 36.4 work_ms   (typical everyday cost)
  confidence : high

No server is running. Nothing touched the network. The pack file carries the schema, the fitted per-resolver costs, and any observed outside hosts: everything a quote needs. The same works from Python:

from costql import PricingPack

pack = PricingPack.load("packs/tmdb_t3.json")
quote = pack.quote('{ movie(id:"27205"){ cast(limit:8){ person{ name } } } }')
quote["price"]         # 50.5384 (safe max, in cost-units)

…and from JavaScript (npm install costql):

import { PricingPack } from "costql";
const pack = await PricingPack.load("packs/tmdb_t3.json");   // or fromObject(await fetch(...).json())
pack.quote('{ movie(id:"27205"){ title } }').price;

The JS package is conformance-tested against the Python engine on every CI run: same pack, same query, same result. See the JS package.

2. Build your first pack against a real API

The Rick & Morty public API needs no keys, so you can do this right now:

pip install 'costql[build]'
costql build --adapter examples/adapters/rickmorty.py:rickmorty_config --out rickmorty.json
costql quote --pack rickmorty.json '{ character(id:"1"){ name episode{ name } } }'

costql build introspects the endpoint, measures a set of clean calibration queries against it (this takes a couple of minutes: you calibrate by measuring), fits the cost model, and writes the pack. From then on, quoting is offline.

Because the Rick & Morty API is a black box we can’t instrument, the pack is T1: wall-clock currency, total-only quotes — and for this API, the fit. On its held-out queries T1 measured ~96% accuracy (case study). The instrumented tiers (T2/T3) see per-resolver and shared work and need one server-side change; whether your API’s shape calls for them is in tier fidelity.

3. Point it at your own API

Two ways in. The fast one: hand the job to a coding agent. Agent-assisted onboarding gives it one skill to follow (skills/costql-adapter) and a prompt that works; the agent probes the endpoint, harvests real IDs, builds, and validates the pack for you.

To write it yourself, an adapter is ~90 lines (where the API is, what real IDs to query with, and a set of curated calibration queries):

cp examples/adapters/rickmorty.py my_api.py   # the minimal T1 template
# edit: endpoint, ids, arg resolution, calibration shapes
costql build --adapter my_api.py:rickmorty_config --out my_api.json

The adapter guide walks through every field. Sanity-check the result with costql validate --pack my_api.json and a few spot quotes.

4. Ship the pack

The pack is a plain file. Vendor it into the app that needs to budget queries; load it with either language’s PricingPack; call quote() before running a query. Every result follows the frozen output contract: price is always present, always a number, always a safe max; cost-units only, never dollars (converting to money is your app’s business).

Cyclic queries (a type re-entering itself through a list edge) are priced as a structural ceiling and flagged confidence: low, never silently billed. See limitations for the full honest list.