Watchful infrastructure for AI apps

See what your app is about to send.

Vectoreyes sits behind your AI endpoint, scrubbing PII, recognising repeated meaning and routing spend before it becomes someone else’s margin.

OpenAI-compatibleEdge-firstBuilt in the UK
01 / observe the requestapi.vectoreyes.app
semantic cache / live trace
edge: eu-west-2
new requestPOST /chat

How do I make a fetch request in TypeScript?

PII scan: clean
92% semantic matchsame meaning
cached answerHIT / 8ms

Use fetch(url, options), then await response.json().

£0.0021 saved · upstream skipped
Change the question. Keep the meaning.
meaning over wording ↗
One endpoint. Three quiet interventions.

The default stack

Most AI apps pay for the same answer twice.

PII travels

User emails, card numbers and identifiers ride along with otherwise harmless prompts.

Bills repeat

Slightly different wording still triggers a full upstream request and a fresh charge.

Keys concentrate

Your gateway key becomes the default payer for every user and every retry.

02 / remember meaning

Same question.
Different words.

The semantic cache looks past phrasing. When the answer already exists, it streams from the edge in milliseconds — no model call, no repeat spend.

semantic cache / live trace
edge: eu-west-2
new requestPOST /chat

How do I make a fetch request in TypeScript?

PII scan: clean
92% semantic matchsame meaning
cached answerHIT / 8ms

Use fetch(url, options), then await response.json().

£0.0021 saved · upstream skipped
Change the question. Keep the meaning.
Cache window

7 days

Response path

edge → stream

Control

cache off, anytime

03 / inspect the route

A small intervention
with a useful memory.

No SDK migration. No new prompt format. Just a more attentive endpoint.

01

scrub

PII stays at the edge

02

embed

meaning becomes searchable

03

check

near-duplicates resolve

04

route

cost follows intent

05

stream

answers arrive at speed

Your codebaseURL: 'https://api.openai.com/v1'
One-line swapbaseURL: 'https://api.vectoreyes.app'

04 / the honest bit

Protective, not magical.

Vectoreyes is operated by Pure Light Ltd, a UK-registered software company. We prefer precise promises to broad ones.

01

PII is sanitised at the edge, in memory, before upstream forwarding.

02

Sanitised prompts and responses are cached at the edge for seven days to serve repeats.

03

Caching is optional. Disable it when your product or policy calls for no retention.

04

Bring your own OpenRouter key on every request, or use a gateway key for the managed route.

Founding access / pre-launch

Be early.
Bring a real endpoint.

We’re inviting a small group of indie developers and SaaS teams to try the gateway against real workloads before general access opens.

No pricing claims. No noise. Just a note when the founding cohort opens. Your email is used only for founding notifications and is deleted on request via hello@vectoreyes.app.

05 / field notes

Questions worth asking.

A gateway should make its boundaries clear. Here are ours.