Toolkit

English · Español

What we actually build with, counted from dependency manifests across our active codebases — not from memory, and not from a list of things we once read about.

Method. Each figure is the number of distinct projects whose package.json or requirements.txt declares that dependency, out of 19 projects that contain code (projects that are documentation or proposals only are excluded). Counting manifests rather than mentions means a technology we integrated once and abandoned does not inflate the number, and a technology we discuss in a design document but never shipped scores zero.

Runtime and framework

	Projects (of 19)
Node.js / TypeScript	13
Express	9
Python	6
FastAPI	3
Cloud Functions Gen2 (functions-framework)	3
Next.js	3
React Native / Expo	1

Google Cloud

	Projects (of 19)
Firestore (firebase-admin)	11
Firestore (google-cloud-firestore, Python)	4
Cloud Storage	5
BigQuery	2
Secret Manager (SDK; most access is via ADC, not the SDK)	1

Compute is Cloud Run and Cloud Functions Gen2 throughout, scale-to-zero by default.

AI / LLM

	Projects (of 19)
Gemini, @google/generative-ai (legacy SDK)	6
Gemini, @google/genai (current SDK)	3
Gemini, google-generativeai (Python)	3

The two JavaScript SDKs coexist because a migration is in progress and we do it incrementally rather than as a single cutover. We state that rather than hide it: a codebase with one SDK everywhere and no migration history is usually a codebase that has only ever been written once.

Data and payments

	Projects (of 19)
MySQL (mysql2, incl. AWS RDS Aurora over a VPC connector)	2
Stripe	1

Third-party platforms

WhatsApp gateways, Meta Graph / Instagram, Telegram (operational alerting), EspoCRM, SendGrid and Resend (transactional email), plus market-data providers in the trading domain. These integrate over HTTP rather than through a vendor SDK in most cases, so they do not appear in a dependency manifest and we do not give them a count here — that would be a number we did not measure.