Bineric Gateway

Every model, agent and MCP tool.One gateway. Your rules.

Put every AI tool, agent and MCP server your organisation uses behind one gateway. Decide who may use what, keep personal data in, cap spend before it runs, and keep a record of all of it.

Hosted by Bineric in the EU, or in your own cloud.

People
Agents
Apps
Gateway
Access
Data rules
Tool approval
Budgets
Audit log
Models
MCP tools
Your servers
support-bot2 names and a phone number maskedMasked
Works withClaude CodeCodex CLICursorContinueOpenAI SDKAnthropic SDKLangGraphn8nBineric AgentsAny MCP server
Organisations that work with BinericMunch MuseumNokutSildreSpoor AISpråkrådetMaanz AI

Your people already use AI, and now agents act on your systems. Security, legal and finance need answers before they can allow it. The gateway gives all three what they ask for.

01 · Access

One sign-in for every AI tool.

Today people sign up with private accounts and pass keys around. With the gateway they log in with their work account, and each team gets the models you have approved for it.

  • Company sign-in through Microsoft Entra ID, Okta, Google Workspace or any OIDC provider.
  • Model access per team, project or application, decided by you.
  • No provider keys on laptops. They stay in the gateway, and people get personal access you can revoke.
  • Offboarding in one place. Remove a person or a team and their access stops.
TeamsPeopleModelsSign-inExample
Teams and model access
TeamAllowed modelsRuns in
Customer service
grp-kundeservice
LynxMistral
EU only
Engineering
grp-utvikling
ClaudeGPTLynx
EU only
Legal
grp-juridisk
Lynx
Own servers
Finance
grp-okonomi
ClaudeMistral
EU only
02 · Agents and MCP

Let agents work. Decide what they may touch.

Agents read your documents, open tickets, write code and send email through MCP servers. The gateway sits between every agent and every tool, so each call is checked against your rules before it runs, and the sensitive ones wait for a person.

An identity for every agent.

Its own access, budget and rate limit, and an off switch in one place.

One registry of MCP servers.

Teams reach only the servers and tools you approve. Servers you have not registered are blocked.

Tools act as the person.

The gateway signs in to each tool on the user’s behalf, so an agent reaches only what its user can.

Allow, ask or block, per tool.

Reading runs on its own. Sending email or making a payment waits for a person to approve it.

Tool results are checked too.

Personal data and secrets in what a tool returns are masked before the model sees them.

Every step on record.

Each model call and tool call, with its arguments and result, in the audit log.

RunsToolsApprovalsExample
invoice-agent, run 2841
  1. modelAllowed
    LynxPlans the task
  2. erp.find_invoicesAllowed
    ERP · MCPReads 14 overdue invoices
  3. crm.get_customerMasked
    CRM · MCP2 personal fields masked in the result
  4. email.sendWaiting
    Outlook · MCPReminder to 14 customers
    ApproveRejectAsks kari.n
  5. bank.create_paymentBlocked
    ERP · MCPRule: payments need finance

Acts as kari.n, Financekr 42 of kr 500 today

ServersToolsSign-inExample
MCP servers
ServerAccessTeams
SharePointRead onlyEveryone
JiraRead and writeEngineering
GitHubRead and writeEngineering
ERPPayments need approvalFinance
mcp.unknown-tools.ioNot registeredNobody
03 · Data

Personal data stays in. Requests stay in the EU.

The hardest question is where the data goes. The gateway checks every request before it leaves, and sends it only to the places you have approved.

  • Personal data masking. Names, identity numbers, emails and phone numbers are masked or blocked before the prompt reaches the model.
  • Region rules. Keep requests to models hosted in the EU, to your own cloud account, or to your own servers.
  • Secret detection. Passwords and API keys pasted into a prompt are stopped.
  • Content is not stored by the gateway unless you ask for it.
Data rulesRegionsTest a promptExample
Data rules for Customer service
What the person wrote

Write a reply to Kari Nordmann, national identity number 010190 12345, at kari@example.no, about her late invoice.

What the model receives

Write a reply to [PERSON], national identity number [ID_NUMBER], at [EMAIL], about her late invoice.

Personal dataMask
Passwords and keysBlock
Processing regionEU only
04 · Spend

See every request, and cap it before it runs.

AI costs are spread over cards, subscriptions and team budgets, and an agent in a loop can spend a month’s budget in an afternoon. The gateway puts it in one place and stops at the limit you set.

  • Usage and cost per person, team, agent, project and model.
  • Budgets per team and per agent, reset every month. At the limit, requests stop.
  • Rate limits, so a runaway script or agent cannot run up the bill.
  • One place to pay. Use your existing cloud agreement, or one invoice from Bineric.
SpendBudgetsRate limitsExample
Spend by team, this month
kr 33 350of kr 53 000 budgeted
Engineeringkr 18 300 / kr 25 000
Customer servicekr 9 100 / kr 15 000
Financekr 4 750 / kr 5 000
Legalkr 1 200 / kr 8 000

Finance is at 95% of its budget. Requests stop at 100%.

05 · Audit

A record you can show.

When the auditor, the data protection officer or the board asks how AI is used in the organisation, you have the answer.

  • Every request recorded: who or which agent, from which tool, to which model, where it ran and what it cost.
  • Every tool call recorded, with what was masked, approved or blocked, and by which rule.
  • Sent to your own logging through OpenTelemetry, next to the rest of your systems.
  • Exports for reviews and reporting.
RequestsPolicy decisionsExportExample
Requests, today
TimeWho and whereModelResult
09:42:10
kari.n
Chat
Lynx
Allowed
09:41:58
ola.h
Coding assistant
Claude
Allowed
09:41:31
kari.n
Chat
Mistral
Masked
09:40:07
invoice-agent
Agent
GPT
Blocked
09:39:44
per.s
Chat
Lynx
Allowed

Delivered as a service. We set it up and run it.

You do not need a platform team. We bring the gateway, the models and the people who know how to connect them, and you choose where it runs.

Hosted by Bineric

We run the gateway for you in the EU.

  • Nothing to install or operate
  • EU-hosted models included
  • Updated and monitored by us

In your own cloud

We install it in your cloud account or data centre.

  • Requests never leave your environment
  • Uses the cloud agreement you already have
  • Run by us, or handed over to your team
  1. 1

    Map

    Which tools, agents and MCP servers your people use, what data they handle, and what your policies and regulators require.

  2. 2

    Set up

    We deploy the gateway, connect your sign-in, tools and logging, and turn your policies into rules.

  3. 3

    Roll out and run

    We connect the tools, help people get started, and operate the gateway or train your team to.

What a security review will ask about

Short answers. We go through the detail with your security and legal teams.

Identity
Sign-in through your identity provider. No shared accounts, and every agent has its own identity.
Residency
Requests go only to the regions and providers you approve.
Personal data
Masked or blocked before it leaves, in prompts and in tool results, by rules you set per team.
Secrets
Provider keys and tool credentials are held in the gateway and never handed to users or agents.
Agent actions
Tools are allowed, approved by a person, or blocked, one by one.
Logging
A record of every request and tool call, sent to your own logging.

Questions

Not answered here? Ask us directly.

Which AI tools work with the gateway?

Tools that let you set where they send their requests, which covers most coding assistants, chat interfaces and agent frameworks, including Claude Code and the Claude desktop app in its enterprise configuration. Your own applications connect with the OpenAI or Anthropic SDKs they already use.

How does the gateway handle MCP servers?

Your MCP servers are registered in the gateway, and assistants and agents reach them through it. You choose which teams may use each server and which of its tools, whether a tool may only read or may also change things, and which actions need a person to approve them. Servers that are not registered cannot be reached.

Can agents run without a person watching?

Yes, within the limits you set. Each agent has its own identity, budget, rate limit and list of tools. Actions you mark as sensitive, such as sending email or making a payment, wait for approval, and you can stop an agent in one place.

Which models can we use?

Models from Anthropic, OpenAI, Google, Mistral and others, through your own AWS, Azure or Google Cloud agreement or through Bineric, as well as open models running on your own servers. You decide which of them each team may use.

Does the gateway store what our people write?

By default it records who made a request, which model or tool answered, how many tokens were used and what it cost, and not the content. If your policy requires the content to be kept, it is stored in your own systems and nowhere else.

We do not have people who can run this. Is that a problem?

No. We set the gateway up, connect it to your systems and operate it for you. If you want to run it yourselves later, we train your team and hand it over.

How does it fit with our data protection work?

The gateway gives you the technical controls a data protection assessment asks for: a known processing location, access control, data minimisation and a record of use. We help you document the setup. The assessment itself remains yours.

What does it cost?

A fixed price for the setup and a monthly price for operation, which depend on where the gateway runs and how many people and agents use it. Model usage is billed at the provider’s price. Get in touch and we will price it for your organisation.

Get in touch.

Tell us which tools and agents your people want to use and what is stopping you. We will show you how the gateway handles it.

Prefer to read first? See our data processing agreement.

  1. 1We reply and set up a first call.
  2. 2We map your tools, agents, data and rules.
  3. 3You get a proposal with a fixed setup price.