Browse documentation
Documentation/Request limits & AI budgets

Request limits & AI budgets

Bound measurement requests and provider calls using your existing database.

These concepts apply to your Python application. Matching and private risk evaluation run in your application-owned TypeScript or Elixir engine; this client handles the HTTP connection. See your setup guide. Code labeled TypeScript describes the engine configuration.

The recommended createDoorman handler includes database-backed limits so browser measurements cannot start unlimited Jev calls. These controls protect the measurement endpoint. Your existing authentication, gateway limits and application policy still protect business actions.

Defaults

The TypeScript createDoorman flow starts with:

Control Default
Measurement requests 600 per minute globally, 60 per supplied account and 30 per supplied session.
Provider calls 60 reserved calls per minute across instances sharing the database and namespace.
Provider concurrency Four reserved calls at once.
Repeated provider failures Pause after three failures; try recovery after 30 seconds.
Measurements in one handler 64 in flight; reuse the handler to make this limit effective.

An identity/risk evaluation reserves two provider calls. Optional cross-device and API activity evaluation share the same budget. This is a call-count limit, not a guaranteed dollar cap. Provider timeouts may still incur charges.

Adjust the limits

import { createDoorman } from "@aarondovturkel/doorman-adapters/node";

const secret = process.env.DOORMAN_IDENTITY_SECRET!;
const namespace = "my-app";
const doorman = createDoorman({
  db,
  secret,
  namespace,
  evaluator: { apiKey: process.env.JEV_API_KEY! },
  evaluatorTimeoutMs: 1200,
  maxInFlightRequests: 32,
  protection: {
    secret,
    namespace,
    requests: { global: 600, account: 60, session: 30, windowMs: 60_000 },
    evaluator: { maxCalls: 60, windowMs: 60_000, maxConcurrent: 4 },
  },
});

These are example operating limits, not throughput targets. Use the same configuration across replicas. Tables are created automatically. No separate protection service or migration is required.

Supply trusted account/session keys when those limits should apply:

const result = await doorman.assess(request, {
  auth: currentUser ? { userId: String(currentUser.id) } : undefined,
  admission: {
    account: currentAccount ? String(currentAccount.id) : undefined,
    session: serverSession.id,
  },
});
return result.response;

The application owns currentUser, currentAccount and serverSession. Never use an arbitrary header, browser JSON value or guessed person ID as a trusted limiter key. Configure database pool and query deadlines separately.

Native Phoenix uses its application’s HTTP/DBConnection admission controls and the configured evaluator protection. Its limits use snake-case options; see the native setup.

Handle limits and failures

  • A measurement request quota returns 429 with Retry-After.
  • A full TypeScript handler or unavailable protection/database storage returns a controlled 503.
  • An exhausted AI budget, timeout or provider failure keeps browser measurement working where storage is available, with riskStatus: "unavailable".

Catch measurement failures in your client so analytics cannot break a successful login. An unavailable zero score does not mean a request is safe. Doorman does not automatically block your checkout, challenge users or show a CAPTCHA.

Add facts your application knows

Pass bounded server counters through riskEvidence, or configure API activity for selected route templates. Trusted edge metadata and optional network reputation are described in identity context. Risk evidence is kept separate from identity matching.

Use your own observability for request counts, latency, errors and budget denials. Do not log full browser payloads, raw IPs, credentials or private candidate identities. Deployment and scale explains what to measure before increasing limits.

Search documentation

Search setup instructions, examples and the API reference.

Local search. No query leaves your browser.