How Routero AI works

Receive the request.

Your app calls api.routero.ai/v1/chat/completions using the OpenAI SDK you already have. We accept any OpenAI-compatible payload — streaming, tools, vision, JSON mode. Beyond chat, embeddings, rerank, image generation, and audio endpoints are exposed under the same API.

app.py

from openai import OpenAI

client = OpenAI(
    base_url="https://api.routero.ai/v1",
    api_key="rt_live_...",
)

resp = client.chat.completions.create(
    model="customer-support",    # or "gpt-5.4", "claude-sonnet-4.6", ...
    messages=[...],
)

Run the policy gate.

Four checks fire in parallel: identity (RBAC, workspace scope), content (prompt-injection score, PII detection), model (is the requested model allowed for this caller), and budget (does this request still fit). Any failure short-circuits with a structured error.

Inbound

openai.chat()

→

Identity · team-finance

Content · 0.02 injection score

Model · customer-support route allowed

Budget · $8,412 / $22,000

→

Decision

ROUTE

Pick the provider.

Routero AI scores eligible providers by health, latency, price, residency, and recent error rate. The chosen provider is called with full streaming pass-through. If it fails or rate-limits, the next eligible provider is tried automatically.

★

Anthropic claude-sonnet-4.6

P50 142ms · $0.003/1k · OK

SELECTED

OpenAI gpt-5.4

P50 178ms · $0.005/1k · 429s

fallback

Gemini 2.0-flash

P50 121ms · $0.002/1k · region-mismatch

ineligible

Mistral large-2

P50 198ms · $0.004/1k · OK

fallback

Account & audit.

Token counts and dollar cost are debited from the budget atomically. The full decision — every check, the chosen provider, the alternatives considered, the response shape — is written to a centralized audit log in real time, queryable and exportable at any moment.

Budget update

team-finance / aug+$0.0124

remaining$13,587.84

Audit entry

{
  "req": "req_8b2f...",
  "user": "u_pa9",
  "route": "customer-support",
  "chose": "anthropic:claude-sonnet-4.6",
  "checks": ["id","content","model","budget"],
  "latency_ms": 142,
  "tokens": { "in": 412, "out": 198 }
}

One API call,
four decisions.

Receive the request.

Run the policy gate.

Pick the provider.

Account & audit.

The Four Building Blocks.

Routes

Policies

Budgets

Audits

Three deployment modes.

Routero AI Cloud

Single-tenant Cloud

Self-hosted (VPC)

See it on your stack.

One API call,four decisions.

Receive the request.

Run the policy gate.

Pick the provider.

Account & audit.

The Four Building Blocks.

Routes

Policies

Budgets

Audits

Three deployment modes.

Routero AI Cloud

Single-tenant Cloud

Self-hosted (VPC)

See it on your stack.

One API call,
four decisions.