About SwitchGate
We're building the boring, trustworthy layer between your product and every AI model - from Mile End, Montréal.
The story
Started with an invoice nobody could explain
SwitchGate began the way most infrastructure companies do: with a spreadsheet and a grudge. Three of us were running AI products on three different provider accounts, and every month ended the same way - a finance person asking what a number meant, and an engineer answering "it depends." Keys lived in shared vaults. An agent once retried itself through a long weekend. The tools to prevent any of this existed at every big company and none of the small ones.
So we built the gateway we wanted to buy: one API in front of every model, provider list prices passed through untouched, one transparent fee at top-up, and governance - budgets, locks, burnable keys, a kill switch - as standard equipment rather than enterprise garnish. Then we went further than the incumbents on the thing we cared most about: your own models, on your own servers, joining the same catalog through the Model Tunnel without your data ever leaving your network.
We host in Montréal on purpose. Canadian privacy law and GDPR adequacy make lawyers relax; Hydro-Québec's grid makes the carbon math pleasant; and the fibre routes between Europe and Ashburn make the physics honest. Also, the bagels are better here, and we will not be taking questions.
Principles
Four things we won't budge on
Pure routing
We never train on your data. The ledger stores math, not conversations. This is the business model, not a feature flag.
Honest numbers
List prices, exact per-call cost, published gateway overhead, and placeholder labels on every metric we haven't earned yet.
Governance for everyone
Budgets and kill switches shouldn't require a sales call. Every safety control ships on the free plan.
Bilingual by design
French isn't a locale file to us - it's routing quality, eval sets and support dans les deux langues, dans le même fil.
At a glance
The facts
| Founded | 2025, Montréal, Québec |
|---|---|
| Product | Multi-model AI gateway: 24 models (plus the auto router), one OpenAI-compatible API |
| Business model | 4.9% fee on credit top-ups; model usage at provider list price |
| Public beta | May 2026 (changelog) |
| Team | Eleven humans, several strong opinions (meet them) |
| Hiring | Yes - four open roles |
| Press | [email protected], subject "Press" |
The rest of the story is on the ledger
Free credits, one line to switch, and a company that would rather show you than tell you.