Sovereign routing · when you want it

The AI router that understands your company's context.

One endpoint. Every model. Dyrout routes each request to the optimal path — by cost, latency, accuracy and data sovereignty.

Built for regulated, enterprise-grade operations
Sovereignty on demand SOC 2 · ISO 27001 ready EU data residency On-prem & self-host
01 — Intelligent routing

One endpoint. Every model. The optimal path, decided per request.

Dyrout sits between your applications and the providers. It reads the context of each request and switches it toward the model that fits — balancing cost, latency, accuracy and where the data is allowed to live.

Incoming request
App · + context
Direct API
POST /v1/route
DYROUT
weighing cost · latency · sovereignty
Claude Opus
GPT-4o
GPT-4o mini
Gemini 2.5
Mistral Large · EU
Llama 3 · self-host
DeepSeek-V3
Command R+
Custom model · sensitive data
optimal path · pin to your locality · −41% cost

Context-aware decisions

Each request is classified by task, risk and priority — then routed to the model that fits, not a fixed default.

50+ managed models, one API

Frontier, open and EU-sovereign models behind a single endpoint — plus your own custom or fine-tuned models for sensitive data.

Sovereignty-first fallbacks

Sensitive data stays on EU-sovereign, on-prem or your custom self-hosted models. Routing respects residency rules as a hard constraint.

Direct API & full governance

Call one OpenAI-compatible endpoint and let Dyrout route. Every decision is logged and explainable — ready for EU AI Act documentation.

02 — Why teams switch

Measurable outcomes, not empty superlatives.

Numbers from production routing across regulated European operations.

−41%
API cost per request
vs single-model baseline
42 ms
p99 latency
SLA < 50 ms held
100%
requests audit-logged
EU AI Act ready
50+
models orchestrated
across 9 providers + custom
03 — Routing console · live demo

See every decision. Try it yourself.

Send a request and watch Dyrout classify it, weigh cost, latency, accuracy and sovereignty, then route it to the optimal model. The metrics below update in real time.

app.dyrout.com/console/routing

Routing dashboard

ABC Corp · Live
Incoming request
— idle —
send a request below
DYROUT
Routed to
optimal path
Send a request
Requests · today
48.302
12.4% vs yesterday
Savings · month
€4.180
31% vs baseline
Avg saved · request
−41%
vs single-model
Latency · p99
42ms
SLA < 50 ms
Latest requests · real-time
No requests yet — send one above ↑
Model distribution
GPT-4o mini38%
Mistral Large (EU)24%
Claude Opus16%
Llama 3 · self-host12%
Gemini Flash10%
Interactive demo · illustrative routing logic & sample data
04 — How we operate

Expert, but not cryptic. Confident, but not arrogant.

We speak the language of technical and business decision-makers — lines, anomalies, KPIs. Not playground jargon.

01

Precise

Every claim is verifiable. Concrete numbers, never empty superlatives.

"−30% API cost per request"
"incredible savings"
02

Sovereign

Europe is our ground. Compliance and data sovereignty are a value, not a constraint.

"native EU AI Act, sovereign cloud"
"compliant with regulations"
03

Operational

We speak the company's language. Real operations, real outcomes.

"your line 3 is in anomaly"
"optimize the workflow"
05 — Pricing

One number. No surprises.

Dyrout charges a flat 5% commission on the tokens you route. You keep paying providers at cost — we add nothing on top but the routing.

5%
commission on token consumption
Pay providers at cost. Model usage is billed through at the provider's own rate — Dyrout never marks it up.
5% only on what you route. No seats, no platform fee, no minimums. If you route nothing, you pay nothing.
Everything included. Sovereignty-aware routing, full audit log, EU AI Act documentation, console and support — all in the 5%.
Routing usually pays for itself. Teams typically cut model spend by far more than 5% by sending each request to the right model.
Estimate your commission
Dyrout commission · 5%
€500
Avg saved · routing
−€3.800
Effective spend
€6.195
Illustrative. Dyrout routes each request to the right model, cutting your underlying provider spend by about 41%; the 5% commission applies to the reduced spend, so the effective cost is well below what you pay today.
Large or regulated estate with on-prem and custom sovereignty needs? Talk to us about volume terms →

The right path, every time.

See how Dyrout routes your real workloads — sovereign, governed, and measurably cheaper. Book a 30-minute technical demo.

Product

Intelligent AI routing, built for the enterprise.

Dyrout sits between your applications and every model provider. It reads the context of each request and routes it to the optimal path — balancing cost, latency, accuracy and data sovereignty.

How routing works

Every request flows through four stages. The whole loop runs in single-digit milliseconds of overhead.

1 · Classify

Each request is scored by task type, risk level, priority and data sensitivity — using lightweight on-path classifiers, not a round-trip to a large model.

2 · Weigh

The router weighs candidate models against your policy: cost ceilings, SLA latency targets, accuracy thresholds and residency constraints.

3 · Route

The request is dispatched to the winning model — frontier, open-weight, EU-sovereign, on-prem or your own custom endpoint.

4 · Log

The decision, inputs, chosen model and rationale are written to an immutable audit log — ready for EU AI Act documentation.

Capabilities

Context-aware decisions
Routing reflects the actual request — a high-risk anomaly goes to a top model, a routine email to a cheap one. No fixed default.
Cost & latency control
Set per-workspace cost ceilings and SLA targets. Dyrout keeps p99 latency under your threshold while minimizing spend.
Sovereignty as a hard constraint
Tag data domains. Sensitive traffic is pinned to EU-sovereign, on-prem or custom self-hosted models — never routed elsewhere.
Custom & fine-tuned models
Register your own endpoints — fine-tuned, distilled or fully private — and let the router send sensitive workloads there with zero data egress.
Direct API, OpenAI-compatible
Point your existing SDK at one endpoint. Dyrout speaks the Chat Completions schema, so migration is a base-URL change.
Fallbacks & retries
Provider outage or rate limit? Dyrout fails over to an equivalent model within policy, transparently.

See routing on your own workloads.

Book a 30-minute technical demo and watch Dyrout route your real traffic — sovereign, governed, measurably cheaper.

Request a demo
Models

50+ managed models. One endpoint. Plus your own.

50+ frontier, open-weight and EU-sovereign models behind a single OpenAI-compatible API — and your own custom or on-prem models for sensitive data, with zero egress.

Managed model catalogue

Dyrout routes across all of these by policy. Add or remove models per workspace; pricing is always passed through at provider cost.

A selection of what Dyrout orchestrates — 50+ models and versions across providers, updated continuously.

Claude Opus 4
Anthropic
frontier
Claude Sonnet 4
Anthropic
balanced
GPT-4o
OpenAI
frontier
GPT-4o mini
OpenAI
fast
o3
OpenAI
reasoning
Gemini 2.5 Pro
Google
frontier
Gemini 2.5 Flash
Google
fast
Mistral Large
Mistral · EU
EU-sovereign
Mistral Small
Mistral · EU
EU-sovereign
Mixtral 8x22B
Mistral · EU
open · EU
Llama 3.3 70B
Meta · self-host
self-host
Llama 3.1 8B
Meta · self-host
self-host
DeepSeek-V3
DeepSeek
open · cost
Qwen 2.5 72B
Alibaba · self-host
self-host
Command R+
Cohere
RAG
Phi-4
Microsoft · self-host
self-host
Jamba 1.5
AI21 · EU
EU-sovereign
Nova Pro
Amazon
balanced
Claude Haiku 4
Anthropic
fast
o4-mini
OpenAI
reasoning
GPT-4.1
OpenAI
frontier
Gemini 2.0 Flash
Google
fast
Mistral Nemo
Mistral · EU
open · EU
Codestral
Mistral · EU
code
Llama 3.1 405B
Meta · self-host
self-host
Qwen 2.5 Coder
Alibaba · self-host
code
DeepSeek-R1
DeepSeek
reasoning
Command R
Cohere
RAG
Phi-4 mini
Microsoft · self-host
self-host
Titan Text
Amazon
balanced
Custom · fine-tuned
Your endpoint
private
Custom · on-prem
Your data centre
private · 0 egress

Custom & sovereign models

For regulated and sensitive workloads, register endpoints Dyrout will treat as first-class routing targets.

On-prem & self-hosted

Run Llama, Qwen, Phi or your own weights inside your data centre. Sensitive requests route there with zero data egress.

Fine-tuned & distilled

Bring a model tuned on your domain. Dyrout sends the requests where your model wins on accuracy or cost.

EU-sovereign providers

Mistral, AI21 Jamba and other EU-hosted models keep regulated traffic inside the Union as a hard constraint.

Bring your own model.

Tell us about your estate and we will map a routing policy across managed, sovereign and custom models.

Request a demo
Security & compliance

Sovereign by design. EU AI Act native.

Dyrout is built for regulated, enterprise-grade operations. Data sovereignty and compliance are core architecture, not add-ons.

EU AI Act, native

Every routing decision is logged, explainable and exportable — the documentation backbone the AI Act expects.

Immutable audit log

Inputs, classification, chosen model and rationale for every request — retained, queryable and exportable for auditors.

Explainable routing

Each decision carries a human-readable reason. No black-box dispatch — you can always show why a model was chosen.

Data residency control

Pin data domains to EU-sovereign or on-prem models. Residency is enforced as a hard constraint, not a preference.

Risk-tiered handling

High-risk tasks get higher-assurance models and stricter logging — aligned with the Act's risk-based approach.

Controls & certifications

Encryption
TLS 1.3 in transit; AES-256 at rest. No request or response content is used to train any model.
Data residency
EU data-residency options across processing and storage. On-prem deployment keeps everything inside your perimeter.
Access & identity
SSO (SAML/OIDC), SCIM provisioning, granular role-based access and per-workspace API keys.
Compliance posture
SOC 2 Type II and ISO 27001 controls; GDPR-aligned data processing; EU AI Act documentation built in.
Retention
Configurable retention and redaction. Sensitive fields can be masked before logging.
EU AI ActSOC 2 Type IIISO 27001GDPREU data residencyOn-premSSO · SCIM

Talk to us about your compliance needs.

We will walk through residency, audit and certification for your environment.

Request a demo
Pricing

One number. No surprises.

Dyrout charges a flat 5% commission on the tokens you route. You keep paying providers at cost — we add nothing on top but the routing.

5%
commission on token consumption
Pay providers at cost. Model usage is billed through at the provider's own rate — Dyrout never marks it up.
5% only on what you route. No seats, no platform fee, no minimums. Route nothing, pay nothing.
Everything included. Sovereignty-aware routing, full audit log, EU AI Act documentation, console, API and support.
Usually pays for itself. Teams typically cut model spend by far more than 5% by routing each request to the right model.
Estimate your commission
Dyrout commission · 5%
€500
Avg saved · routing
−€3.800
Effective spend
€6.195
Illustrative. Dyrout routes each request to the right model, cutting your underlying provider spend by about 41%; the 5% commission applies to the reduced spend, so the effective cost is well below what you pay today.
Large or regulated estate with on-prem and custom sovereignty needs? Talk to us about volume terms →

What's included

Routing & models

50+ managed models, custom & on-prem endpoints, sovereignty-aware routing, fallbacks and retries.

Governance

Immutable audit log, explainable decisions, EU AI Act documentation, role-based access and SSO.

Platform

Routing console, direct OpenAI-compatible API, usage analytics, cost & latency dashboards, support.

Start routing today.

No seats, no minimums — just 5% on the tokens you route.

Request a demo
Documentation

Build on Dyrout in minutes.

One OpenAI-compatible endpoint. Point your existing SDK at Dyrout, add an optional routing policy, and let it choose the optimal model per request.

Quickstart

Three steps: get a key, set the base URL, send a request. Migration from OpenAI is a one-line change.

Route a request
# Route a request — Dyrout picks the optimal model
curl https://api.dyrout.com/v1/route \
  -H "Authorization: Bearer $DYROUT_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role":"user","content":"Classify this contract clause..."}],
    "policy": {"max_risk":"high","residency":"eu","optimize":"cost"}
  }'
# OpenAI-compatible — just change the base_url
from openai import OpenAI

client = OpenAI(
    base_url="https://api.dyrout.com/v1",
    api_key="$DYROUT_KEY",
)

resp = client.chat.completions.create(
    model="dyrout-auto",            # let Dyrout route
    messages=[{"role":"user","content":"Summarize this report..."}],
    extra_body={"policy":{"residency":"eu","optimize":"cost"}},
)
print(resp.choices[0].message.content)
print(resp.model)                          # which model Dyrout chose
// OpenAI-compatible Node SDK
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.dyrout.com/v1",
  apiKey: process.env.DYROUT_KEY,
});

const resp = await client.chat.completions.create({
  model: "dyrout-auto",
  messages: [{ role: "user", content: "Draft an email..." }],
  // @ts-ignore — Dyrout routing policy
  policy: { residency: "eu", optimize: "cost" },
});
console.log(resp.choices[0].message.content);

Use model: "dyrout-auto" to let Dyrout route, or pass a specific model id to pin it. The policy object is optional and overrides workspace defaults.

Authentication

All requests use a bearer token. Keys are scoped per workspace and can be rotated at any time.

Header
Authorization: Bearer $DYROUT_KEY

Create and rotate keys in the console under Settings → API keys. Keys never expire automatically but can be revoked instantly.

API reference

A compact, OpenAI-compatible surface plus Dyrout routing extensions.

POST/v1/routeRoute & run a request
POST/v1/chat/completionsOpenAI-compatible alias
GET/v1/modelsList available models
GET/v1/usageUsage & commission
GET/v1/auditExport routing audit log

Routing policy parameters

FieldTypeDescription
optimizeenumcost · latency · accuracy · balanced — what the router prioritizes.
residencyenumeu · on_prem · any — hard constraint on where data may be processed.
max_riskenumlow · medium · high — minimum assurance tier for the chosen model.
max_costnumberPer-request ceiling in EUR. Router excludes models that would exceed it.
sla_msnumberTarget p99 latency in milliseconds.
allow / denystring[]Explicit allow/deny lists of model ids.

Response shape

Response
{
  "id": "route_9f2a...",
  "model": "mistral-large-eu",          // chosen by Dyrout
  "routing": {
    "reason": "eu residency + cost-optimal",
    "risk": "medium",
    "saved_vs_baseline": "0.41",
    "latency_ms": 42
  },
  "choices": [{ "message": { "role":"assistant", "content":"..." } }],
  "usage": { "prompt_tokens": 812, "completion_tokens": 240 }
}

The routing object explains the decision — model chosen, reason, risk tier, estimated saving vs a single-model baseline and measured latency. This is what gets written to your audit log.

SDKs & tooling

OpenAI-compatible

Any OpenAI SDK works — Python, Node, Go, Java, .NET. Change the base URL and you are routing.

Streaming

Server-sent events on /v1/route and the chat alias, identical to the OpenAI streaming format.

Webhooks

Subscribe to usage thresholds, policy violations and audit events for your SIEM.

Get your API key.

Request access and start routing across every model from one endpoint.

Request a demo
Company

Built in Europe, for European operations.

Dyrout is the enterprise AI routing infrastructure by BEXT S.r.l. — engineered for organizations that need control over cost, latency and data sovereignty.

Who we are

BEXT S.r.l. is a European software company based in Paratico (BS), Italy. We build infrastructure that lets regulated organizations adopt AI without giving up control of their data or their costs.

Precise

Every claim is verifiable. Concrete numbers, never empty superlatives.

Sovereign

Europe is our ground. Compliance and data sovereignty are a value, not a constraint.

Operational

We speak the language of real operations — lines, anomalies, KPIs — not playground jargon.

Work with us.

Whether you are evaluating routing or ready to deploy, we would like to hear about your operations.

Request a demo