Sign inGet started

Routes

Steer every model call your agents make.

An LLM router for your agents: one URL, every model, with your own keys and rules. Open up a new world of observability and dynamic optimization.

Get startedSee how it works
Works withAnthropicOpenAIOpenRouterFireworksTogetherGroqCerebrasDeepSeekxAIGemini
steerz.recurz.ailivesample data
The Steerz Live view for an example team: 2.4 requests a second now, 9 in flight, none failing; requests per second over the last minute; 4,320 requests in the last hour, p50 3.3 s, $634 spent in the last 24 hours at list price; and where calls went, from three regions to five models, each with its requests and cost.
Every call, from every region, liveEvery call priced as it landsSpend at list price, by model

01 · Connect

Change the base URL. Keep everything else.

Claude Code, Codex and every major SDK work unchanged, streaming and tools included.

# Claude Code
export ANTHROPIC_BASE_URL=https://steerz.recurz.ai
export ANTHROPIC_AUTH_TOKEN=stz_northwind_…
claude
steerz.recurz.ai/connectsample data
The Steerz Connect page: a box to confirm a key works, then snippets for Claude Code, the Claude Agent SDK, Codex CLI, the Anthropic and OpenAI SDKs, the Vercel AI SDK and curl.
A snippet for every client

02 · Your keys

Your keys, sealed to your org.

Each provider key is encrypted with your org and the provider sealed in, so it can’t be opened for anyone else.

steerz.recurz.ai/connect#modelssample data
The Steerz model list with input, cache read, cache write and output prices per million tokens, and the provider keys panel explaining how keys are sealed.
Encrypted before it’s stored

03 · Rules and budgets

Every call priced, routed and capped.

Rules swap models, clamp effort or drop fields. Each call reserves its worst case against your budget, so a burst can’t overrun it.

steerz.recurz.ai/livesample data
The Steerz live tail: each call’s time, model, the rule that routed it, region, outcome, input and output tokens, duration and cost, newest first.
The rule behind every callReal cost, per call

How it works

Your call, out and back.

Each call runs through a small router compiled to WebAssembly at the edge. It never holds your prompt longer than the call. If you turn on response caching, matching answers are kept for up to 7 days, scoped to your org.

  1. Rule. A rule can rewrite the model, clamp effort or drop fields.
  2. Reserve. The call’s worst case is held against your budget.
  3. Call. The provider is called with your org’s own key.
  4. Meter. The stream is counted as it passes: input, cache reads and writes, output.
  5. Settle. The charge settles at the real cost, and the rest of the hold goes back.

Better together

Steerz sits in front of every model call

TracezSteerz plus Tracez unlocks Insights++: why the cache missed, the exact prompt on every turn. HandzSpend and loops, watched live across the fleet. VerdictzProves which model wins on your own work before a rule sends traffic to it.

07 · Routes

Which model should do what?

Steerz answers it for every model call your agents make, wherever they run.

Early access

Every model. Your rules.

Get startedEmail the foundersSee the API

Available now: the edge router, your own keys for every major provider, rules, spend limits and the live view. Coming: adaptive routing, Insights++ in Tracez, your own small models as routes.