The inference gateway for AI products
Alectura sits between your product and its model providers. Every request gets the policy its customer requires: where it’s processed, which models see it, what’s kept, what’s redacted, and what happens when an account turns abusive.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://us.gateway.alecturalabs.com/openai/v1",
api_key=os.environ["ALECTURA_API_KEY"],
)
response = client.responses.create(
model="gpt-5-mini",
input="Say hello in French.",
extra_headers={
"Alectura-Organization": "initech",
"Alectura-User": "user_42",
"Alectura-Plan": "pro",
},
)
print(response.output_text)
import os
from anthropic import Anthropic
client = Anthropic(
base_url="https://us.gateway.alecturalabs.com/anthropic",
api_key=os.environ["ALECTURA_API_KEY"],
)
message = client.messages.create(
model="claude-haiku-4-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Say hello in French."}],
extra_headers={
"Alectura-Organization": "initech",
"Alectura-User": "user_42",
"Alectura-Plan": "pro",
},
)
print(message.content[0].text)
import os
from google import genai
client = genai.Client(
api_key=os.environ["ALECTURA_API_KEY"],
http_options={
"base_url": "https://us.gateway.alecturalabs.com/gemini",
"headers": {
"Alectura-Organization": "initech",
"Alectura-User": "user_42",
"Alectura-Plan": "pro",
},
},
)
response = client.models.generate_content(
model="gemini-3.5-flash",
contents="Say hello in French.",
)
print(response.text)
Why
Neither is about model quality.
01 · Control
Enterprise buyers want a say in how their data meets a model: which region processes it, which models may see it, whether anything is kept, and what is redacted. Each asks for something different, and most products can give one answer for everyone.
Alectura gives each of your customers its own policy.
02 · Abuse
Accounts on free and cheap plans get farmed, wrapped in adapters and resold as someone else’s tokens, and much of what flows through is content you don’t want your name on. Signup checks can’t see it: it shows only across the requests themselves.
Alectura watches users across requests, and stops them.
Both sit at the same place in your architecture: where your servers call a model provider. That’s where Alectura runs.
How it works
Every request runs the same pipeline, compiled from the policies that match it. Everything before the provider happens in memory, so the provider is the only network hop, and the budget for it all is 3 ms at the median for bodies up to 256 KB.
On the way back, the steps run in reverse over the stream: pseudonyms are restored as text arrives, and a stream an injection check stops ends in the provider’s own error format, so your error handling already works.
Policy per customer
A policy matches requests by organization, user or plan, and every policy that matches applies. Guardrails, limits and retention take the strictest value any of them sets. Regions, models and connections are lists that start from the request: a policy can keep only some of one or remove some, and add a region or a connection, but no policy’s add brings back what another kept out.
{
"match": {},
"settings": {
"routing": {
"regions": { "add": { "global": {} } }
},
"guardrails": {
"secrets": { "action": "redact" }
},
"retention": { "content": "30d" }
}
}
{
"match": { "plan": ["free", "trial"] },
"settings": {
"routing": {
"models": { "only": ["claude-haiku-4-5"] }
},
"limits": {
"spend": [
{ "per": "user", "period": "day", "amount": "2.00" }
]
},
"abuse": { "suspend_at": 0.9 }
}
}
{
"match": { "organization": "globex" },
"settings": {
"routing": {
"regions": {
"add": { "eu": { "rank": 1 } },
"only": ["eu"]
},
"models": { "only": ["claude-sonnet-5-5"] }
},
"retention": { "content": "none" },
"guardrails": {
"personal_data": { "action": "pseudonymize" }
}
}
}
Effective policyA request from Globex, on the enterprise plan, to a US environment
euThe request starts with us, Baseline adds global and Globex adds eu, then Globex keeps only euclaude-sonnet-5-5Globex keeps only this onenoneGlobex, stricter than Baseline’s 30 daysredactBaselinepseudonymizeGlobexWhat it does
On the record
Each request records the policies that matched it at their versions, what each detector found and what it did, and what nothing could inspect. A verdict never claims more than its coverage.
Findings
emailpersonal_datapseudonymized ×2phonepersonal_datapseudonymized ×1aws_access_keysecretsredacted ×1Coverage
systemsecrets · personal_datausersecrets · personal_data · prompt_injectiontool_resultsecrets · personal_data · prompt_injectionattachmentimages_uninspectedRegions
Each region runs a full copy of Alectura. An environment is pinned to one when it’s created: its requests are routed, processed and stored there, and leave only for a provider its policy allows.
US
AWS us-east-1, N. Virginia
us.gateway.alecturalabs.comEU
AWS eu-west-1, Ireland
eu.gateway.alecturalabs.comAU
AWS ap-southeast-2, Sydney
au.gateway.alecturalabs.comYour keys, your contracts
Requests go to your own provider accounts, on your own keys, so the prices, zero-retention terms and regional commitments you negotiated stay in force.
Create an environment, connect a provider, and change two lines.