The provider for
agentic engineers.
Every model at full native power, without committing to one provider or juggling three or four accounts.
A single-provider setup gives your agents real capability, but it locks you to one company's models and one account. Switchboard gives you all of it. Every model behind one API, spend caps for unsupervised runs, and per-token usage on every request. No provider accounts to open, no keys to paste.
Get started
Personal use
One balance and every model for your own work, straight from your coding agent.
pi extension quickstart →For your team
One prepaid balance for your whole team, with spend caps and per-person usage. No provider accounts.
Team quickstart →Build with it
Ship the SDK, meter every end user per token, attribute usage per key.
Pick your SDK →Routes to
Compatible with
Reach for Switchboard when
Keep using whatever works. But the moment a plan, a model, a forge, or an IDE cannot flex to the problem in front of you, that is when you reach for Switchboard.
You have hit your usage limit.
Your agents still have work to do and the flat plan has stopped. Top up your balance and keep going on any model, at each provider's exact rate.
You run reviews and automations all day.
Put the volume on the cheapest model that does the job, spawn sub-agents for the heavy passes, and let router rules fire your review loops automatically.
You are tired of being locked in.
A plan or a model, a forge or an IDE. Each wants to be the only tool you use, and each ships an agent that only works inside its own walls. Switchboard stays flexible, so your agents reach every model and every program you work in.
You want to build your own tools and automations.
Ship on the SDK and every call is metered, capped, and billed like the rest, whether it runs for you or for the customers of what you build.
How does Switchboard help?
Routing
One request shape reaches every live model in the catalog. Change models by changing the model id. Switchboard speaks each provider's native format, so nothing is lost. Streaming, tool calls, and caching all work at full capability.
Metering
Every token measured as it streams and itemized at provider-exact rates. Input, output, and reasoning are all counted. Cached-prompt discounts included. You know what any request cost the moment it finishes.
Control
Spend caps, rate limits, and model policy. All enforced before the call, not in your code. Every agent run is a controlled box, an isolated sandbox with its own budget. Its sub-agent trees are bounded in depth and width, so nothing spirals into infinite delegation or sibling explosions. Permissions are gated at the server, so the agent cannot change its own settings or grant itself a tool. You turn agents loose. The walls hold.
Research
Capabilities, configuration, and per-token prices for every model. Human-verified in Linecard before they go live. Compare the new frontier model against your current one before routing a single request to it.
Analytics
Usage by model, by user, and over time. Typed error records in each user timeline. You see who used what and what it cost. When a request failed, you see why.
Key management
Switchboard mints and manages keys for your application and every end user. Nothing ships in a binary, and nothing gets pasted. Rotation happens over the wire, with no client rebuild and no downtime.
Attribution
Every provider on one balance. Every token attributed to the agent, member, or end user that spent it. Your own runs or a thousand end users, you can see exactly where inference went and what it cost, down to the token.
Automation
Spawn independent sub-agents from inside your agent session, on any model in the catalog. Fan the grunt work out to Kimi or DeepSeek while the main task keeps going. Rules you set at the router steer a session when a matching tool call fires. Review loops and follow-ups run on your logic and your model choice, on the same balance.
Built for how you work
One platform, from your first agent to a product serving thousands.
Solo developer
Every model from one balance, so you stop opening an account per provider. A meter on every task shows what your agent runs cost, and spend caps keep an unsupervised run inside a budget you set.
Startup team
One prepaid balance for everyone, with caps and model policy set at the router so autonomous runs stay bounded without anyone babysitting them. Every token is attributed to the member that spent it.
Building a product
Start by metering your own agents. When your product starts calling models for its customers, meter them the same way. A managed key per end user, spend caps per end user, and provider-exact rates on every call. Inference grows with your product as your customers use it, one SDK and per-token usage per end user, ten or ten thousand.
Our promise
Full native power,
or we don't serve it.
We take inference management seriously. You will never get less from a model through Switchboard than you would going direct. If it is on our list, it runs at full capability, in its own native format. We handle the maintenance and details that make that work; you change a model id and ship.
Get started →import Switchboard
guard let apiKey = ProcessInfo.processInfo.environment["SWITCHBOARD_API_KEY"] else {
fatalError("Set SWITCHBOARD_API_KEY in the environment")
}
let client = Client(apiKey: apiKey)
let response = try await client.inference(Inference.Request(
model: "claude-sonnet-5",
messages: [.user("Hello")],
user: "your_user_id"
))
print(response.choices.first?.message.content ?? "")curl https://switchboard.valni.app/v1/switchboard/inference \
-H "Authorization: Bearer $SWITCHBOARD_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-sonnet-5",
"messages": [{"role": "user", "content": "Hello"}],
"user": "your_user_id"
}'Pricing
$19.99 per seat, plus inference at passthrough + 4.9%.
During the alpha, every seat is free. Top up credit and pay only for inference at each provider's exact rate.
Switchboard FAQs
Can I keep using my existing coding subscription?
Yes. Switchboard is a separate provider you run alongside anything else you already use. It gives your agents every model at full native power on one balance, with spend caps and metering. Most teams run the bulk of their agent work through it. Keep another subscription for the model you drive by hand, or do not. Switchboard does not depend on it.
How is Switchboard different from OpenRouter or LiteLLM?
OpenRouter and LiteLLM are gateways. They unify the API. Switchboard is built for engineers who work through agents. It gives you per-agent and per-member usage attribution, rate limits and model policy enforced before the call, and local models behind the same interface. Per-token accounting spans your whole team. It also mints and manages every key, so nothing gets pasted around.
Which request formats are supported?
The formats you already use. Switchboard speaks each provider's native format, lossless, and every model is auto-configured from Linecard to run with its full capability set. Provider-specific options pass through untouched when you need them.
Which models can I route to?
Every live model from an enabled provider in Linecard is routable through Switchboard. The registry tracks pricing, capabilities, and configuration per model. You can pick by feature, like tool calling or vision or caching, not just by name. The live list is always one call away, and browsable in Linecard.
How does billing work?
A $19.99 per-seat monthly platform fee, waived during the alpha, plus a 4.9% routing fee applied once when you top up credit. Usage draws down your balance at each provider's exact rate. Start with standard rate limits. Higher limits come with history or by application.
What powers the routing
See pricing, capabilities, and configuration for every model
Linecard is the model registry behind Switchboard. Track what each model can do, what it costs per token, and the configuration it performs best with.