AI Gateway

One entry point to every model

Your scripts and agents talk to Supervise; Supervise talks to the models. Bring your own keys at no extra cost, or use ours to call every model in one line.

Request a demo See the platform

Your tools

Ticket triage

Weekly digest

invoice_export.py

Invoice reminders

+ 38 agents and scripts

supervise AI Gateway

12,480calls today

Models

OpenAI

Anthropic

Mistral

Google

Meta, DeepSeek…

01 · Your keys or ours

Your keys or ours, and never in the code again

Plug in your own API keys: you pay your providers directly, at no extra cost. Or use ours: every available model is one line away, with no account to open at each provider.

Your keys, no extra cost

One key per provider, owned by the company, in the vault. You pay on the terms of your own contracts.

Our keys, every model

Text, voice, image, video: every available model in a single call, on one usage-based bill.

No more secrets in the code

Cybervise blocks any key written in plain text; the Gateway makes it unnecessary.

API keys3 personal keys in code

invoice_reminders.pyScript · Inès B.

OPENAI_KEY = "sk-proj-…4f2a"client = gateway("gpt-4.1") Personal key

Ticket triageAgent · Camille M.

ANTHROPIC_KEY = "sk-ant-…91bc"client = gateway("claude-sonnet") Personal key

export_crm.pyScript · Sofiane R.

MISTRAL_KEY = "mst-…7d1e"client = gateway("mistral-large") Personal key

Company vault

OpenAI••••

Anthropic••••

Mistral••••

Google••••

Your keys: no extra cost · or ours: every model

02 · Automatic failover

A provider goes down, your tools keep running

When a model stops responding, the Gateway instantly switches to the fallback model you chose, then returns to the primary as soon as it is back.

A fallback for every use

You choose the fallback model, tool by tool.

No call lost

Failover happens call by call, with no interruption for your teams.

Automatic return

The primary model takes over again as soon as it is back. Every switch is traced.

Automatic failoverTicket triage

PrimaryOpenAI · gpt-4.1Operational

FallbackAnthropic · claude-sonnetStandby

Calls per secondOpenAI Anthropic

100%calls succeeded

0calls lost

—failover time

03 · Call log

Every call, in real time, down to the cent

No more personal keys scattered across tools and bills aggregated by provider. Every call is visible, explained, attributed.

One line per call

Time, script, model, latency, tokens, cost. Exportable, audit-ready.

Today’s cost, this month’s cap

You watch consumption move and get alerted before the overrun.

Failover, traced too

When a fallback model takes over, the log says so.

Call log Live

All OpenAI Anthropic Mistral Google

TimeToolModelCost

14:02:40Enrichmentmistral-small€0.001

14:02:31Invoice remindersgpt-4.1-mini€0.002

14:02:23Invoice exportmistral-large€0.006

14:02:19Weekly digestclaude-sonnet€0.041

14:02:17Ticket triagegpt-4.1€0.018

Spent today

€61.4051% of €120 cap

04 · Allowed models

Each team only calls the models you allow

A simple grid: teams on one side, models on the other. A call to a model that isn’t allowed is refused, logged, and its owner is notified.

Rules per workspace

Finance, Support, HR: each workspace has its own list of allowed models.

Refused, not just flagged

A call outside the rules doesn’t go out. It is traced in the log.

Change models without touching the code

Tools call the Gateway; the rule decides which model answers.

Allowed modelsper workspace

GPT-4.1ClaudeMistralGemini Support Finance Sales HR Legal

AllowedNot allowed

Last call · FinanceAccepted

invoice_reminders.py→gpt-4.1

Logged · €0.002

05 · Budgets and caps

A budget per team, and not a euro more

Each workspace has its monthly cap. The owner is warned before it is exceeded, and they decide whether to raise it.

An alert before the overrun

At 80% of the budget, the workspace owner is notified.

A cap that holds

At the cap, new calls wait for approval.

The cost of every call

Every euro is tied to its tool, its team and its task.

AI budgetsOctober

Support /

Finance /

Sales /

Ops /

Finance · 80% of budget

Inès B., the workspace owner, has been notified.

Live in 5 minutesOutlives its authorBuilt-in SSOGDPR and AI ActAutomatic failoverScheduled, loggedZero duplicatesYour keys or oursLive in 5 minutesOutlives its authorBuilt-in SSOGDPR and AI ActAutomatic failoverScheduled, loggedZero duplicatesYour keys or ours

Analysed on every deploymentHosted in France and EuropeShared catalogCapped budgetsEvery call loggedSecrets in the vaultThe models you chooseAlerts and automatic retriesAnalysed on every deploymentHosted in France and EuropeShared catalogCapped budgetsEvery call loggedSecrets in the vaultThe models you chooseAlerts and automatic retries