Requesty: AI Gateway & LLM Router for 600+ Models

Q: What is Requesty?

An AI gateway between your app and 600+ models from every major provider. Change your base URL to router.requesty.ai and get intelligent routing, fallbacks, cost optimisation, auto-caching, governance and observability.

Q: How do I integrate?

One line of code: client = OpenAI(base_url='https://router.requesty.ai/v1', api_key='your-key'). Works with all major SDKs.

Q: How does it reduce costs?

Smart routing to cheaper equivalent models, auto-caching, automatic fallback from expensive providers, per-user spending limits, and real-time cost analytics.

Q: Does it work with Cursor, Cline and Continue?

Yes. Native OAuth integrations, any model, and no rate limit from us. The only ceilings are the ones you set on the key.

Q: How does pricing work?

5% markup on model costs. All features included. Enterprise plans available with volume discounts.

Q: Can I use my own API keys?

Yes. Bring your own keys for any provider while getting Requesty's routing and observability. Or use our unified key.

Dashboard

Your AI ops, at a glance

Heatmaps, cost breakdowns and cache gauges across all your AI providers, on a dashboard that refreshes as the requests land.

requestyDashboard / Overview7d30d90d30D

Total requests

143.2K

18.7%

Total cost

$1,247

12.3%

Cached input

87.3%

4.2 pts

Success rate

99.85%

30 days, this account

Latency distribution24H UTC

<50ms

<200ms

<500ms

<1s

<2s

>2s

00061218

<50ms3.6%

<200ms32.4%

<500ms43.1%

<1s17.2%

<2s3.4%

Share of requests, low to high. Peak load at 14:00 UTC.

Auto-cachingPROVIDER PREFIX

87.3%Cached input

2.63B

Tokens cached

78.6%

Off input

$1,029

Saved

Daily cost by model7D

M

T

W

T

F

S

opus-4.6gpt-5.4gemini-3.6-flashdeepseek-v4-prokimi-k3

By model30D

opus-4.6Anthropic34.2%$427

gpt-5.4OpenAI25.5%$318

gemini-3.6-flashGoogle19.6%$244

deepseek-v4-proDeepSeek12.7%$158

kimi-k3Moonshot8.0%$100

Plug and play

Integrate in a minute

Point your base URL at Requesty and keep everything else. One line of your code changes.

OpenAI-compatible API, works with any SDK
Your prompts stay yours: switch model or leave with one line
Automatic failover and load balancing included

Read the quickstart

from openai import OpenAI
client = OpenAI(
    api_key="YOUR_REQUESTY_KEY",
    base_url="https://router.requesty.ai/v1",
)
resp = client.chat.completions.create(
    model="anthropic/claude-fable-5",
    messages=[{"role": "user", "content": "Hello"}],
)

import OpenAI from "openai";
const client = new OpenAI({
    apiKey: process.env.REQUESTY_API_KEY,
    baseURL: "https://router.requesty.ai/v1",
});
const resp = await client.chat.completions.create({
    model: "anthropic/claude-fable-5",
    messages: [{ role: "user", content: "Hello" }],
});

curl https://router.requesty.ai/v1/chat/completions \
  -H "Authorization: Bearer $REQUESTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-fable-5",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

EU customers swap the host for router.eu.requesty.ai. Same API, same key, and the request terminates in Frankfurt against EU-hosted model endpoints.

That one base URL reaches:

Anthropic
OpenAI
Google
Mistral
DeepSeek
Moonshot AI
Z.ai
xAI
Alibaba
AWS Bedrock
Azure OpenAI
Meta
and every other major provider

01Route

Intelligent infrastructure

Where a request goes, what it is allowed to cost, and what happens when the provider behind it fails.

Region routingONE ENDPOINT

Gateway regions and what the routing policy on the key does with a request from each
Region	Serves	Policy
eu-central-1pinned	EU, Frankfurt	EU models only
us-east-1	US, Virginia	nearest healthy
ap-southeast-1	APAC, Singapore	nearest healthy

Request from an EU key for a US-only modelrefused, not forwarded

A pin is not a preference. A call that would leave eu-central-1 returns an error rather than crossing the border.

Team policySUPPORT

Modelsopus-4.6gpt-5.4gemini-3.6-flash3 of 600+

Endpointseu-central-1eu-west-1EU only

Spending limits by scope
Scope	Limit	Used
Organisation acme-inc	$2,000/mo	62%
Team platform	$900/mo	57%
Team support	$500/mo	65%
Team research	$260/mo	91% near cap
Team growth	$300/mo	58%
Key ci-eval-runner	$60/mo	22%

A call from support for deepseek-v4-pro, or to us-east-1, is refused at the gateway: not billed, not forwarded, logged with the rule that stopped it.

02See

Real-time observability

Cost, latency and usage across every provider, current enough that you find a problem while it is still small.

Cost attributionLAST 30D

30d spend

$1,247

Prev 30d

$1,422

Change

-12.3%

Per 1K req

$8.71

The six largest spend lines this month, by team, key and model
Team	Key	Model	Spend
platform	prod-api	opus-4.6	$101
support	triage-svc	opus-4.6	$79
platform	prod-api	gpt-5.4	$76
support	triage-svc	gpt-5.4	$59
platform	prod-api	gemini-3.6-flash	$58
support	triage-svc	gemini-3.6-flash	$45

03GovernSecurity

Governance and guardrails

Sensitive data scrubbed before the outbound call, policy enforced per team, user and key.

PII detection and scrubbing

Personal data redacted before it reaches the model. Emails, phone numbers, national IDs and cards, among eight entity types.

PII redaction8 ENTITY TYPES

Your application sends

Customer sarah.chen@northwind.co called about invoice 8841. Callback +44 7700 900412, card ending 4242.

Scrubbed at the gateway, 3 of 8 types matched

The model receives

Customer [EMAIL] called about invoice 8841. Callback [PHONE], card ending [CARD].

Detection runs in your region. 8 types: email, phone, credit card, national ID, IBAN, IP address, postal address and date of birth.

Content guardrails

Enforce content policies, block prompt injections and filter harmful outputs, on the way in and on the way out.

GuardrailsLAST 24H

Prompt injectionBlocking4,773 checked

Content policyBlocking4,773 checked

PII redactionRedacting318 scrubbed

Topic filterBlocking4,773 checked

Toxicity shieldBlocking12 blocked

Rate limiterThrottling0 throttled

Every guardrail is enforcing. Two of them had something to act on in the last day.

Team management

Role-based access with Owner, Admin, Billing, Developer and Viewer roles. Set per-team budgets, model allowlists and usage quotas.

Roles and permissions37 MEMBERS

Permissions by role
Role	Keys	Budgets	Billing
Owner	Yes	Yes	Yes
Admin	Yes	Yes	No
Billing	No	Yes	Yes
Developer	Yes	No	No
Viewer	No	No	No

Five roles, plus per-team budgets, model allowlists and usage quotas on top.

Audit logs

Complete audit trail of every action. Track who did what, when and from where. Export logs for compliance and review.

Audit logEXPORTABLE

Recent audit events
Time	Action	Origin
14:22	j.okafor raised the research budget to $260/mo	SSO
14:09	ci-eval-runner key rotated	API
13:54	p.nowak pinned acme-prod to eu-central-1	SSO
13:31	d.mensah added gpt-5.4 to the support allowlist	SSO
12:47	Guardrail toxicity-shield blocked 4 requests	SYSTEM

Every action, actor, timestamp and origin. Streamed to your SIEM or exported as JSON.

In production

What it looks like when the gateway stops being your problem

We evaluated over 7 different gateways and ended up with Requesty for their enterprise readiness and configurability.

Arkady Landes

Senior Cloud DevOps Manager, ZoomInfo

Read the ZoomInfo story

2.45M

Requests, 100% EU

77B

Tokens, ~31K per request

Requesty became our default AI control plane. We control where data goes, which models run on which tasks, and what we spend, from one place. For a regulated business, that is exactly the layer we needed.

Rechtsanwalt Michael Friedmann

Vorstand, CLIO (Chief Legal Innovation Officer), anwalt.de services AG

Read the anwalt.de story

100%

Traffic, EU residency

~50%

Tokens served from cache

Requesty sits in the critical path of every phone call we answer, and we simply stopped thinking about it. Calls are faster, costs are lower and the data stays in Europe.

Ignacio Ruisánchez

Founder and CEO, NotarBot

Read the NotarBot story

Also in productionAll case studies

99.95%

success rate or better, every month so far, Mozart AI

40%

lower AI cost per meeting, Amberscript

3x

faster rollout of new models, LOUPZ

330+

football clubs on the platform, Soccerment

Review and residency

Built to survive procurement

The questions security and legal ask, answered before they ask them, including the one where the answer is not yet.

In progress

SOC 2 Type II

Observation period under way with an independent auditor. Controls in place and documented; the target date is on the trust page.

Contractual

GDPR, DPA on request

An Article 28 DPA, signed on request at any spend, with sub-processors listed.

Regional

EU residency

Frankfurt, on AWS eu-central-1, via router.eu.requesty.ai.

Retention

Zero data retention, pinnable

Pin your traffic to endpoints that retain nothing. 131 of the EU model endpoints qualify.

Pricing

One number, and it sits on top of the model cost

No per-seat pricing, no feature gates, no minimum. You pay for what your application spends, plus 5%.

Pay as you go

5%

markup on model cost

Every feature included: routing, caching, failover, guardrails, analytics and audit logs. No seat fees, no feature tiers.

See pricing

0%Bring your own keys

on your own provider contracts

Keep the pricing you have negotiated with each provider and still get routing, limits and observability across all of them.

How BYOK works

VolumeEnterprise

discounts and committed spend

SSO, SCIM, private regions, a signed DPA and a named contact. Procurement paperwork handled by people, not a form.

Speak to a founder

Questions and answers

What is Requesty?

An AI gateway between your app and 600+ models from every major provider. Change your base URL to router.requesty.ai and get intelligent routing, fallbacks, cost optimisation, auto-caching, governance and observability.

How do I integrate?

One line of code: client = OpenAI(base_url='https://router.requesty.ai/v1', api_key='your-key'). Works with all major SDKs.

How does it reduce costs?

Smart routing to cheaper equivalent models, auto-caching, automatic fallback from expensive providers, per-user spending limits, and real-time cost analytics.

Does it work with Cursor, Cline and Continue?

Yes. Native OAuth integrations, any model, and no rate limit from us. The only ceilings are the ones you set on the key.

How does pricing work?