Trade-offs · 2026-11 · 1 min read
Integrations live on your machine, not in a cloud
Whether an outside tool needs an account is not a business decision — it is decided by whose machine holds the data. One rule, a whole class of mistakes gone.
Most tools sort integrations by what the tool does: a citation manager, a version control system, a search backend. We sort them by something simpler — whose machine holds the data. Yours means no account and no network; someone else’s server means a connector.
The Zotero test
Zotero is the cleanest example. The desktop app exposes a local HTTP endpoint at 127.0.0.1:23119 — the same port the Word plugin uses. Not one byte leaves the machine and no OAuth is involved. So Textrix treats Zotero as a built-in local integration: read-only access to items, notes and PDF paths, with citations resolved through Better BibTeX citation keys.
Zotero also offers a web API at api.zotero.org backed by its cloud library. We do not use it. It would put accounts, tokens and a server between you and your own library — the exact arrangement local-first exists to prevent.
Local processes need no permission
The same rule covers git, python and language servers: they live in layer two, reached through local ports, CLIs or inter-process communication. No authorisation flow, works offline — and if the process is not running, the integration reports that honestly instead of pretending otherwise.
What the rule rules out
Anything whose data only exists on someone else’s server needs an account by definition — that is layer three, and it is what connectors are for. It is a smaller set than most tools pretend, and keeping it small is a feature: fewer places for your research data to pass through.
Keep reading
Two more articles from the same trade-off series.
Still in development
Not yet available for download. Join the waitlist and you will be invited first when it opens.
Email-collection mode: submitting opens a pre-filled message in your mail client.
