SecuriX

Enterprise LLM + MCP Gateway

The control plane for your organization’s AI traffic.

SecuriX sits in front of every LLM and MCP server your teams touch — enforcing SSO, budgets, virtual keys, and DLP policy before a token leaves your network. Runs on-prem or in your own VPC with Docker.

Admin consoleSSO connected · Okta
Monthly budget$42,300 / $60,000

Virtual keys

128 / 6 teams

DLP this month

214 redactions

Same policies, either way
On-premYour VPC · Docker

The hidden cost of AI

“You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful.”
Satya NadellaChairman & CEO, Microsoft · The Reverse Information Paradox, July 2026
Cost 01 · Money

The token bill

Seats and API usage spread across teams, keys, and providers — usually seen for the first time on the invoice.

Per-team budgets, virtual keys, and spend attributed to a real identity.

Cost 02 · Know-how

Everything you feed the model

Prompts, documents, agent tool calls, and corrections — your institutional knowledge, leaving with every request.

DLP on every prompt and completion, policy on every MCP tool call, and a full audit trail.

Before you roll out another ChatGPT or Claude subscription, know what it costs and what it learns. SecuriX is the agent access security broker, LLM gateway, and MCP gateway that gives you full clarity on your company’s AI usage.

See how SecuriX helps

Access control

Single sign-on, enforced at the gateway.

Connect Okta, Azure AD, or any SAML/OIDC provider. Every request is tied to a real identity — no shared API keys floating around in Slack.

  • SAML 2.0 & OIDC
  • SCIM provisioning
  • Role-based access per team
Identity providers
Okta · SAML 2.0connected
Azure AD · OIDCconnected
Google Workspaceavailable
128 users provisioned via SCIM

Cost control

Set a budget. Never blow through it by accident.

Cap spend per user, team, or project. SecuriX downgrades or blocks requests automatically when a limit is hit — before the invoice surprises anyone.

  • Per-team and per-project caps
  • Automatic model downgrade at threshold
  • Real-time spend alerts
Spend by team · this month
Engineering$18,200 / $25,000
Support$6,100 / $10,000
Growth$8,000 / $8,000
Growth capped at 100% — auto-downgraded to a cheaper model

Credential hygiene

Generate keys your real provider credentials never touch.

Every team, app, or environment gets its own scoped virtual key — rotate or revoke instantly without touching the underlying OpenAI, Anthropic, or Azure credentials.

  • Scoped to model, rate, and team
  • Instant revoke, no redeploy
  • Full usage attribution per key
Virtual keys128 active
sk-vx-prod-checkoutgpt-4o, claude-sonnet-5 · 500 req/min
revoke
sk-vx-staging-support-botclaude-haiku · 100 req/min
revoke
sk-vx-sales-copilotgpt-4o-mini · 50 req/min
revoke

Data protection

Catch sensitive data before it leaves your network.

SecuriX inspects prompts and completions for PII, secrets, and policy violations in real time — redact, block, or just log, per policy.

  • Built-in PII & secret detectors
  • Redact, block, or log per policy
  • Full audit trail per request
Policy log live

Redacted SSN pattern in prompt · checkout-bot

Redacted AWS access key in completion · sales-copilot

Blocked customer record export · support-bot

214 redactions this month · 0 leaked

Tool governance

Granular policy on every MCP tool call — not just the prompt.

SecuriX runs as the MCP gateway between your users and your MCP servers, plus a growing set of first-party integrations — so every tool call passes through policy, not just the prompt. Conditions on the request itself — who's asking, what they're touching, what it's worth — decide whether it runs.

  • Gateway for any MCP server, not just first-party
  • Attribute-based conditions, not just per-team access
  • Allow, block, or require approval per action
MCP tool policiesattribute-based
calendar.create_eventattendees: internal only
allowed
github.merge_pull_requesttarget: main, approvals: 0
blocked
stripe.issue_refundamount > $1,000
needs approval
gdrive.share_fileclassification: confidential, recipient: external
blocked

Governed RAG

Ask your company's docs — but only the ones you're allowed to see.

Upload documents and SecuriX indexes them for retrieval-augmented answers. Scope a doc to a team — Finance, say — and only that team's questions ever surface its chunks, no matter who else asks.

  • Per-team and per-user document scoping
  • Access enforced in the retrieval index, not the UI
  • Same identity, same permissions as everywhere else
Knowledge base
Priya · Finance3 sources found

“What was our Q3 burn rate?”

Marcus · Support0 sources visible

“What was our Q3 burn rate?”

Q3-Board-Deck.pdf is scoped to Finance — the index respects it, not just the UI

Deployment

Runs on-prem or in your VPC. Never ours.

Deploy SecuriX in your data center or your own cloud account with a single docker run. Prompts, responses, and audit logs stay inside your boundary — we never host your traffic.

  • One Docker image, on-prem or VPC
  • Your data never touches our servers
  • Hosted sandbox available for demos and POCs
Deployment
On-prem
  • docker run securix/gateway
  • Your data center
  • Same admin console
Your VPC
  • AWS, Azure, or GCP
  • Your cloud account
  • Same admin console

The DLP layer

Three actions. Your real data never leaves your infrastructure.

Every prompt and completion passes through SecuriX’s detectors for PII and secrets. Each policy decides what happens next.

Audit

Flag it, let it through.

The prompt reaches the model unchanged. SecuriX records what it found, who sent it, and where — so you can see exposure before you enforce anything.

Summarise the dispute for Emily Carter, card ending 4421.

flagged · 2 findings · sent as-is
Redact

Swap it, restore it.

Real values are replaced with realistic synthetic ones before the request leaves. The model still gets a coherent prompt — just not your data.

Summarise the dispute for Laura Bennett, card ending 8830.

redacted · 2 values swapped
Block

Stop it at the gateway.

The request never reaches the provider. The user gets a clear policy message instead of a completion, and the attempt is logged.

Export every customer record with SSN for the audit.

blocked · policy: no-bulk-pii
Redact · round triprealsynthetic
  1. 01
    Your userSends a prompt

    Email Emily Carter at emily@acme.com about her refund.

  2. 02
    SecuriXSwaps real for synthetic

    Emily Carter → Laura Bennettemily@acme.com → laura@example.net

  3. 03
    Model providerOnly ever sees fakes

    Hi Laura, your refund is on its way. Reply to laura@example.net with any questions.

  4. 04
    SecuriX → your userRestores the originals

    Hi Emily, your refund is on its way. Reply to emily@acme.com with any questions.

SecuriX runs on-prem or in your VPC, so the real-to-synthetic mapping never leaves your infrastructure. The provider sees a coherent conversation; your user sees their real data.

How it works

From SSO to your first governed request in under a day.

01

Connect identity & providers

Point SecuriX at your Identity Provider for SSO (Okta, Azure AD, or any SAML/OIDC provider) and the model or MCP providers your teams already use. No client-side code changes.

02

Define policy

Set budgets, virtual keys, DLP rules, and attribute-based MCP tool policy — per team, per user, or per request attribute. As granular as your org needs.

03

Route through the gateway

Point client apps at your SecuriX endpoint instead of the provider directly. Deploy on-prem or in your VPC with a single docker run.

04

Monitor & audit

Every request, tool call, and knowledge-base query lands in one console — spend, redactions, and policy decisions, all attributable to a real identity.

See SecuriX in front of your own traffic.

Tell us where to reach you and we'll set up a walkthrough with your identity provider, your policies, and your budget structure.

Request a demo

We'll reach out within 2 business days.