Every model.One API.Full control.
Route every request through one governed gateway, with budgets, locks, and tracing included.
from openai import OpenAI
client = OpenAI(
base_url="https://api.switchgate.ai/v1", # the only change
api_key="sg-live-…",
)
r = client.chat.completions.create(
model="anthropic/claude-sonnet-4-6",
messages=[{"role": "user", "content": "Bonjour!"}],
)Model catalog
100+ models, one endpoint
Anthropic, OpenAI, Mistral, Groq, Bedrock… at provider list price, pass-through, never marked up.
No models match your search.
Placeholder data for design purposes; the live catalog ships with the API.
All models → Discover → Rankings → Benchmarks →Governance & security built in
Control your OpenRouter bill can't buy
The only gateway that routes French intelligently, and the only one that treats governance and security as a feature, not an add-on.
Per-user locks and budgets
One place for every model your people call. Lock which models each user or key can reach, cap spend before it happens, and see it all from a single console — whether traffic comes from an app, a script, an agent or the chat.
Bring your own model
Install the device agent on your server and it tunnels your model into the gateway. Your teams keep the API they already use; your weights keep the network they already trust.
Keys assigned to people
Every key belongs to a named user or service: per-person budgets, per-person revocation, per-person spend on the ledger. Burnable prepaid keys cover contractors and classrooms.
Runaway agent kill switch
A velocity breaker stops looping agents the instant they blow through your thresholds — before the invoice does the talking, not after.
Exact cost per call
Full request tracing with the exact cost of every call across every provider, in one ledger — and semantic cache hits land at exactly 0 credits.
Measured, not promised
Cross-provider retry turns individual outages into one measured effective success rate — the number that matters when a provider stumbles at 2 a.m.
Routing that speaks French
French-aware routing detects francophone traffic at the gateway and sends it to the models that actually handle it well — a routing feature, not a locale file. Built in Québec, tested on Québec.
Pricing
Same provider prices. Lower fees.
Model usage is billed at provider list price, pass-through. The only difference is the fee when you top up.
| Top-up comparison | SwitchGate AI | OpenRouter |
|---|---|---|
| Platform fee on top-ups | 4.9% | 5.5% |
| Minimum fee | $0.50 | $0.80 |
| Paid credits | Never expire | Expire after 12 months |
| Provider prices | List price, pass-through | List price, pass-through |
Token economy mode
Semantic cache hits are billed 0 credits. Repeated prompts stop costing money.
70%
Teams typically save 30–70% on cache-friendly workloads.
- Canadian hosting 🇨🇦
- Loi 25 / GDPR-friendly
- Live status page
- No training on your data, pure routing
From the blog
Notes from the gateway
Product releases, engineering notes and benchmark write-ups.
Introducing the runaway agent kill switch
A velocity breaker that stops looping agents the instant they blow through your thresholds, before the invoice does the talking.
Aug 6, 2026 EngineeringHow French-aware routing actually works
Language detection at the gateway, a francophone quality index per model, and what happened when we routed Québec traffic with it.
Jul 22, 2026 BenchmarksGateBench Airline: 110 models, one rubric
Our multi-turn tool-calling benchmark is now reproducible end to end. Here is the harness, the rubric and the surprises.
Jul 9, 2026Change one line. Keep full control.
Point your OpenAI-compatible SDK at SwitchGate and keep every budget, lock, and trace.