Skip to main content
Introducing The Quantum Infrastructure Cloud Learn more

InfraGuard · Private AI access control

Private AI, with control over every system it touches

InfraGuard is software you install on your own servers, between your AI agents, your people and your infrastructure. Each call is allowed, sent for approval or denied by your policy, and written to a record you can verify. Your models, your data and your access rules stay on your machines.

Works with the systems you run

  • Linux
  • VMware vSphere
  • Kubernetes
  • PostgreSQL
  • OpenTelemetry
  • Hugging Face
  • Ollama
  • Keycloak

What you get

Software you install on your own servers, licensed by what it protects

InfraGuard is not a cloud service and not a chatbot. It is a gate you run next to your systems: every person, agent and application that wants to reach them goes through it, and it decides each call.

Who calls

People

Sign in with your SSO: OIDC, Active Directory, SCIM

AI agents

Tools over MCP or REST, or a hook in the runtime you use

Apps that call models

Any OpenAI-compatible client, pointed at the gateway

InfraGuard

What you install: one image, on your servers

  • ConsoleInventory, access, approvals and audit

  • Policy decisionAllow, require approval or deny, with a decision id

  • Tool gatewayTyped tools for agents, over REST and MCP

  • Model gatewayGrants with calls per minute and tokens per day

  • AuditA hash-chained record you verify offline

Docker Compose Helm Licence file, checked offline

Your systems

Linux servers and VMs

On VMware, Proxmox or bare metal, over SSH

Kubernetes

Pods, logs, events and restarts

PostgreSQL

Logins that expire by themselves

Logs and events

Searched inside the caller’s role

Your models

Open-weight, on your GPUs

Every call from the left is decided by InfraGuard before it reaches the right.

01 · Agree the scope

A paid pilot

An InfraGuard engineer agrees with you which systems, agents and models the pilot covers, and you receive the offline bundle.

Request a pilot

02 · Install

On your machines

Verify the bundle and start it with Docker Compose or Helm. Without a licence file it runs as Community: 25 protected resources and 3 AI agents, with people not counted.

How to install

03 · Run it

An annual licence

Growth, Business or Enterprise. Your limits are written in a licence file that your installation checks offline. A larger estate means a new file, with no restart.

Talk to us
Protected resource
A server, database, cluster or other system you register with InfraGuard.
Governed AI agent
Each AI agent or model workload held to your policy.
Not counted
People, logs and events. Collectors are not limited by their own number.

Prices are agreed with you and are not published on this site.

Private AI with access control

Let agents search and act across your estate. Only what your policy names will run.

Allow, approve or deny
one decision per call, with an id and the policy version
Denied by default
a call no policy matches is refused
Typed tools only
natural language never becomes a privileged command
No call home
InfraGuard makes no call to SoftQuantus; licences are checked offline

Search with access control

One search across logs, events and inventory, inside your role

People and agents ask the same question. InfraGuard answers from what their tenant and role may read, and nothing older than the retention you set.

Logs · OpenTelemetryKubernetes eventsInventoryPostgreSQL, read-only
What failed on payments-api since 09:00?Search

Results for your role

  1. 01logpayments-api-7c9fupstream timeout calling ledger-db after 5 s09:14
  2. 02k8s_eventpayments-apiBack-off restarting failed container09:16
  3. 03inventoryledger-db-01PostgreSQL 16 · Linux VM · owner: paymentsnow

Scope of this answer

Tenant
ops-eu
Role
operator
Retention
30 days

Events of other tenants are never returned, and nothing older than the retention is served.

Events per hour

06:0012:00

Illustration: names, identifiers and numbers are examples.

See it decide

Agents reach information under control

An incident, a private model and a database login: three short replays of what InfraGuard decides, line by line, and why.

An operations agent investigates a slow payments service for the engineer on call.

Caller
ops-agent for the on-call engineer
Systems
Kubernetes, PostgreSQL, Linux

    A replay with example names, identifiers and limits. The tools, the decisions and the rules are InfraGuard’s.

    What it does

    Control of your models, your data and your access

    Copper streaks of light with three glass tiles joined by lines: a person, a shield with a tick and a ledger.

    Access control

    Agents and people reach systems only through policy

    Every tool call says who asked, for whom and what. A delegated agent gets the stricter of its own policy and the person’s. A write or privileged action runs only when a policy names it, and an approval runs it once.

    Enterprise grade
    Violet streaks of light with three glass tiles joined by lines: a spark, a shield and a server rack.

    Private models

    Your LLMs, on your GPUs, behind one gateway

    Serve open-weight models from Hugging Face on your own machines. Each workload needs a grant, with calls per minute, tokens per day and an output limit. Prompts are not stored, and no service holds the model’s key.

    Private models
    Green streaks of light with three glass tiles joined by lines: a magnifying glass, a shield and a ledger.

    Search and observe

    Search what runs, without handing over the keys

    Agents search the inventory, read logs, metrics and Kubernetes events, and query databases read-only. Logs and metrics arrive over OpenTelemetry, and an OpenSearch mirror keeps one role per tenant.

    On-premises

    Private LLMs

    Hugging Face models, with the InfraGuard layer in front

    Choose an open-weight model on the Hugging Face Hub, download it once, and serve it on your own GPUs with vLLM or llama.cpp. With the reference egress rules, InfraGuard is the only way in: each workload, agent and chat gets a grant with its own limits, prompts are not stored, and every call is recorded.

    • A grant per workload, model by model
    • Calls per minute, tokens per day, output size
    • Known secret formats redacted before the prompt leaves
    • Deprecation dates and a circuit breaker
    • Streaming answers, counted as they pass
    • 429 with Retry-After when a budget is spent
    How model governance works
    1. Workloads, agents and the chat

      Each with its own token, grant and limits

    2. InfraGuard model gateway

      Grant, budget, redaction and a record of every call

    3. Your model server

      vLLM or llama.cpp on your GPUs; InfraGuard holds its key

    4. Hugging Face Hub

      Open weights, downloaded once into your network

    A call through the gateway
    POST /v1/llm/chat/completions
    Authorization: Bearer <workload token>
    
    { "model": "vllm/qwen3-8b",
      "messages": [{ "role": "user", "content": "…" }] }

    Recorded from the product

    Watch it run

    Four recordings from the public tutorials: install, enrol a server, govern a model, check the audit.

    A recording of the tutorial on docs.infraguard.cloud.Follow the tutorial

    Your estate

    Built for the systems your company already runs

    Each tile says how InfraGuard reaches the system today. Nothing on this list is a promise of a future connector.

    Servers and virtual machines

    • Linux servers

      Red Hat Enterprise Linux, Ubuntu, Debian, SUSE

      SSH command profiles: typed parameters, never a free shell line

    • VMware vSphere

      ESXi hosts and their guests

      Linux guests through SSH command profiles

    • Microsoft Hyper-V

      Windows Server virtualisation

      Linux guests through SSH command profiles

    • Proxmox VE and KVM

      Open-source hypervisors

      Linux guests through SSH command profiles

    • People on servers

      Engineers and administrators

      Short-lived SSH certificates issued by InfraGuard

    Containers and data

    • Kubernetes

      OpenShift, Rancher and other distributions expose the same API

      Pods, logs, events, restarts and discovery; tested on k3s

    • PostgreSQL

      Operational databases

      Read-only queries, writes under policy, temporary logins that expire

    • PostgreSQL in your cloud

      Azure Database for PostgreSQL, Amazon RDS, Google Cloud SQL

      The same PostgreSQL connector; not yet verified on these managed services

    Identity

    • Microsoft Active Directory

      Over LDAPS

      People, groups and nested membership, synchronised on a schedule

    • Keycloak and OIDC

      Single sign-on

      Signed access tokens; tested end to end with Keycloak 26.4

    • SCIM 2.0 and SAML

      Provisioning and federation

      SCIM provisioning; SAML through an identity broker

    Logs, monitoring and security operations

    • OpenTelemetry

      OTLP over HTTP

      Logs and metrics in, each stored under its sender’s credential

    • OpenSearch

      Version 3.9

      A search mirror with one role per tenant

    • Your SIEM

      Security operations

      The audit chain by cursor, and signed webhook alerts

    Private AI

    • Hugging Face

      Open-weight models from the Hub

      Download once, serve on your GPUs, govern every call

    • vLLM

      Model serving

      The serving engine of the Private AI bundle

    • llama.cpp and Ollama

      Local model servers

      Any OpenAI-compatible or Ollama endpoint behind the gateway

    • LibreChat

      Chat for employees

      Private AI bundle: chat, model and sign-in on your machines

    • MCP and REST agents

      Agent frameworks and assistants

      One tool gateway over MCP streamable HTTP and REST

    Where it runs

    • Docker Compose

      A single host

      The reference installation

    • Helm

      Your Kubernetes platform

      A chart with reference egress rules

    • Air-gapped networks

      No internet at all

      An offline bundle, verified locally, with an offline licence

    Every integration, with its tests

    Not a chatbot over your servers

    A gate in front of them

    Give AI the reach your teams need, and keep the decision with your policy, your approvers and your audit.

    Request a pilot

    Why InfraGuard

    Six things that set it apart

    1. 01

      A decision before the action

      InfraGuard decides before a call reaches the system. It is a gate, not an alert after the fact.

    2. 02

      No free-text commands

      Tools take typed arguments and command profiles. What a model writes is never run as a shell line.

    3. 03

      Agents hold no credentials

      Agents act through InfraGuard. People get SSH certificates and database logins that expire on their own.

    4. 04

      The model key stays inside

      Services call the model through the gateway and never see its key, so a leaked token cannot reach the model elsewhere.

    5. 05

      A record you can check

      Every decision joins a tamper-evident hash chain. You verify it, and your SIEM reads it by cursor.

    6. 06

      Runs where your data is

      In your data centre, your private cloud or an air-gapped network, from an offline bundle you verify yourself.

    Get started

    From the bundle to the first refused call

    InfraGuard is delivered as an offline bundle through a pilot. Verifying and loading it needs no network and no registry, and the running service makes no call to SoftQuantus.

    From the install guide
    cd infraguard-offline-<commit>
    ./verify-release.sh --key infraguard-release.pub --commit <commit>
    # VERIFIED: infraguard-api-<commit>.tar.gz ...
    docker load -i infraguard-api-<commit>.tar.gz
    # Loaded image: infraguard-api:<commit>
    1. 01

      Read the install guide

      Requirements, configuration and the first steps are public. Read them before you talk to us, so the pilot starts on your questions.

      Install guide
    2. 02

      Request a pilot

      Tell us what your agents and people should reach. We agree a scope with you and send an offline bundle and, where the pilot needs one, a licence.

      Request a pilot
    3. 03

      Verify, load, start

      Check the bundle with its own verification script before anything runs, then start InfraGuard with Docker Compose or the Helm chart, next to your PostgreSQL and identity provider.

      Release verification
    4. 04

      Protect the first resource

      Register one cluster, database or server, write the first policy and connect an agent over MCP. The tutorials walk through each step.

      Tutorials

    Questions

    Before a pilot

    Is InfraGuard a chatbot?

    No. InfraGuard is the control layer between AI and your systems. Your agents, assistants and people keep the tools they use; InfraGuard decides what each of them may reach and records every decision. The optional Private AI bundle adds a chat for employees, governed the same way.

    Which models can we use?

    Any model service with an OpenAI-compatible chat completions route, or an Ollama endpoint, that you run or choose. For a fully private setup, serve open-weight models from Hugging Face with vLLM or llama.cpp on your own GPUs.

    Does any of our data leave our network?

    InfraGuard makes no call to SoftQuantus, and its licence is checked offline. Data goes only to the model services, identity provider and search cluster you configure. With models on your own GPUs, nothing has to leave at all.

    Which systems can agents reach today?

    Kubernetes, PostgreSQL and Linux hosts over SSH, including Linux virtual machines on VMware vSphere, Hyper-V, Proxmox and KVM, plus the inventory, logs and metrics InfraGuard collects. The integrations page of the documentation lists each one with its tests.

    What does it cost?

    Community runs without a licence file, for up to 25 protected resources and 3 governed AI agents. Growth, Business and Enterprise are set by a licence agreed with us. We start with a paid pilot against an agreed scope, and we do not quote prices on this site.

    Request an enterprise pilot

    Tell us what your agents and people need to reach, and where InfraGuard would run. It takes about two minutes.

    I work in *
    About you
    Your estate

    The film and pictures on this page are SoftQuantus’s own, painted from noise and line work; the clips are recordings of the InfraGuard tutorials. Product names belong to their owners and show where InfraGuard fits, not a partnership or an endorsement.