Axum guide · Rust SDK

Axum rate limiting,as one Tower layer.

Add ReqKeyLayer to your Router. Every request is checked against the caller’s key, charged against their credits, and held to their plan’s limit before your handler runs.

Start building freeRust SDK docs
  • Keys issued per customer
  • Credits that refill
  • Limits per plan, not per process

Official SDKs with drop-in middleware for the stack you already run

  • Python
  • Node.js
  • Go
  • Rust
  • PHP
  • .NET
  • Java
API requests validated
100M+
Average key validation
<5ms
Average analytics ingest
<5ms
Check and log, end to end
<10ms

The usual Axum rate limiter counts requests. It doesn’t know your customers.

Most Axum services reach for tower_governor, a Tower layer over the governor crate’s GCRA limiter, keyed by peer IP by default with custom key extractors available.

With tower_governortoday
  • Limiter state lives in the process — each replica has its own
  • Keyed by IP address by default, not by the customer who is paying
  • No API keys: issuing, hashing, scoping, and revoking them is still yours to build
  • No credit balance: a request can’t cost 5 on one route and 1 on another, or refill each month
  • One limit for everyone — no per-plan limits for Free, Pro, and Enterprise customers
  • No per-customer usage log to answer “why was I blocked?” or bill against
With ReqKey in Axumone middleware
  • API keys issued per customer, prefixed, hashed, and revocable from the dashboard
  • A credit balance per customer — price each route, refill every hour, day, week, or month
  • Rate limits set per plan, shared by all of a customer’s keys, from 1 second to 24 hours
  • The same count on every worker, instance, and region — no Redis to run
  • Over the limit? A 429 with Retry-After, and no credits charged
  • Every request logged per customer — status, latency, endpoint — in under 5ms
Handlers take Extension<VerificationResult> to read the verdict. See the code

Axum in three steps, one of them code.

ReqKey runs inside your API, not in front of it. Your server asks one question per request and gets an answer in under 5ms.

  1. 1

    Install the SDK

    cargo add reqkey --features axum

  2. 2

    Set your project key

    Copy it from the dashboard into REQKEY_PROJECT_KEY. It stays on your server.

  3. 3

    Add the Axum middleware

    Every request is checked, charged, and logged before your handler runs.

main.rscargo add reqkey --features axum
use axum::{extract::Extension, routing::post, Json, Router};
use reqkey::{
    axum::ReqKeyLayer,
    middleware::{KeyLocation, Middleware, MiddlewareConfig},
    Client, VerificationResult,
};
use serde_json::{json, Value};
 
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = Client::from_env()?;
    let config = MiddlewareConfig::builder("api_payments")
        .key_location(KeyLocation::header("X-Startup-Key"))
        .exclude_path("/health")
        .build()?;
 
    let app = Router::new()
        .route("/payments", post(create_payment))
        .layer(ReqKeyLayer::new(Middleware::new(client, config)));
 
    let listener = tokio::net::TcpListener::bind("127.0.0.1:3000").await?;
    axum::serve(listener, app).await?;
    Ok(())
}
 
async fn create_payment(Extension(decision): Extension<VerificationResult>) -> Json<Value> {
    Json(json!({ "credits_remaining": decision.credits_remaining }))
}
Also in Rust: Actix Web, Rocket, WarpFull Rust reference

Every request gets one of these answers

  • 200Valid key, credits charged — your handler runs
  • 402Out of credits
  • 403Key disabled, or not allowed on this API
  • 429Over the rate limit — no credits charged

Where ReqKey sits

Your customer
Your APIReqKey, under 5ms
Your handler

Responses go straight back to your customer. ReqKey sees the key check and the log line, nothing else.

Credits

Charge each route what it costs you.

A lookup can cost 1 credit and a render 5, from the same balance. Excluded paths are never validated, charged, or recorded. In Rust:

Per-route credits · Rust
// Exact paths or trailing-* prefixes: never validated, charged, or recorded
.exclude_path("/health")
.exclude_path("/docs/*")
 
// Or decide per request with a predicate
.should_protect(|request| request.path.starts_with("/api/"))
 
// Charge different endpoints differently (non-negative u64)
.credit_cost_resolver(|request| {
    Ok(if request.method == Method::POST { 5 } else { 1 })
})

Rate limits

Set limits on plans, not in code.

Your Axum code never hard-codes a number. Each plan carries its credits, refill, and rate limit; moving a customer to Pro changes all three with no deploy.

PlanCreditsRate limit
Free1,000 / month5 req / s
Pro50,000 / month50 req / s
Scale1,000,000 / month500 req / s
429Over the limit, a call is answered with Retry-After and costs nothing — no credits and no quota.

Example plans. You name them and pick the numbers.

Notes for Axum teams hit in production.

A regular Tower layer

ReqKeyLayer composes with TraceLayer, CorsLayer, and the rest of tower-http in whatever order you need.

Typed decision

Extension<VerificationResult> hands the verdict to any handler that asks for it — credits remaining, consumer, request ID.

Safe body capture

Response bodies are recorded only when textual, about 4 KB or smaller, uncompressed, and of known size — so streaming responses are never buffered.

Axum rate limiting: the questions teams ask.

Something else? Ask the team or read the docs.

  • No. It uses your Tokio runtime through the async Client, created once with Client::from_env().

  • Yes — apply .layer(...) to a nested Router, or use exclude_path for public paths.

  • On their plan or on the customer (consumer) in ReqKey — a number of requests per window from 1 second to 24 hours, shared by all of that customer’s keys. Your Axum code never hard-codes a limit, so upgrading a customer is a dashboard change, not a deploy.

  • A check averages under 5ms. ReqKey runs inside your app as middleware, not as a gateway in front of it, so responses go straight back to your customer.

  • You choose: fail closed and answer 503, or fail open and let requests through. Invalid keys are denied either way, and a validation is never retried, so no one is charged twice.

More Rust guides.

Other languages

Ship API keys in Axumin five minutes.

Free for your first 5 million requests every month. No card, no gateway, no rewrite.

Start building freeRead the docs