LLM gateway — bring your own provider keys
Keep your AI features up when a model provider goes down.
Switchyard gives your apps one OpenAI-compatible endpoint. Behind it, an alias tries your primary model first and falls back to the next provider on errors, timeouts or rate limits — using your own keys, with per-key limits and logs.
- Free: 1,000 requests a month
- No markup on tokens
- Prompts not stored by default
All 5 steps in this demo (example data, recorded in the real app)
How it changes your week
Before and after Switchyard.
The manual way
- Provider SDKs and keys scattered across services
- Hand-written retry-and-fallback code in every app, slightly different each time
- One provider outage or 429 becomes your outage
- No single place to see which model answered, how fast, and at what token count
With Switchyard
- Change the base URL and key once; any OpenAI SDK keeps working
- Aliases like “smart” try up to five provider/model targets in order
- On 5xx, 429, timeouts or unreachable providers, the next target answers
- Every request logged with provider, model, status, latency, tokens and tries
In the demo the primary provider is deliberately unreachable (a local test setup), so you can see the fallback happen.
What you get
Three things you get on day one.
Automatic fallback
An alias is an ordered list of provider and model targets; your code only knows the alias.
How: moves on when a target returns 5xx or 429, times out (you set the per-attempt timeout) or can’t be reached; 4xx errors come back to you unchanged.
Two lines to switch
Keep your OpenAI SDK, LangChain or curl calls; change the base URL and key.
How: OpenAI-compatible chat completions with streaming; Anthropic requests and streams are translated to the OpenAI format.
See every request, not the content
Usage per alias and per key, and a log entry for every call.
How: model, provider, status, latency, tokens and number of tries; prompt and response content only if an admin turns it on.
Try it now — no sign-up
See the change in your code.
Pick your client and an alias name. This is the whole migration: a new base URL, a gateway key, and the alias as the model name.
- Works with existing OpenAI SDKs
- Streaming (SSE) supported
- Per-key rate limits
Everything included
The rest of the toolkit.
Keys stay with us, encrypted
Provider keys are encrypted at rest and never sent back to your apps.
Per-key limits
Requests per minute per gateway key, with a workspace default.
Usage by alias and key
Requests and tokens this month, broken down so you know who used what.
CSV logs
Download request logs as CSV; the whole workspace as JSON.
Direct addressing
Skip aliases when you want: call provider/model directly.
Team access
Invite teammates with their own login and role.
Pricing
Start free. Upgrade when it pays for itself.
Free: 1,000 requests and 1M tokens a month, 2 providers, 5 aliases, 7-day logs — no card needed. Pro is $49 a month for 200,000 requests. You pay your providers directly; no overage charges from us.
Free
$0 /month
- 1,000 requests / month
- 1M tokens / month
- 2 provider connections
- 5 model aliases
- 7-day request logs
- Bring your own provider keys
Pro
$49 /month
- 200,000 requests / month
- 200M tokens / month
- 20 provider connections
- 200 model aliases
- 30-day request logs
- Everything in Free
Prices exclude VAT. UAE customers pay 5% VAT. All prices are in US dollars. Questions? WhatsApp us
FAQ
Before you sign up.
Do you charge for tokens?
Do you store my prompts?
Which providers work?
What happens at my plan limit?
Can I cancel, export or delete?
Are prices including VAT?
What isn’t included?
One endpoint. Every provider. Fewer outages.
Connect two providers, create an alias, change your base URL. Free, no card.