Skip to main content
Governed infrastructure as code

Change infrastructure fast, without taking it on faith.

Your team describes what they need. Intentric turns it into a specific change, checked against your rules and waiting for a person to approve. Then it is applied with OpenTofu or Pulumi, confirmed against the request, and kept that way — with a record of every step.

No AI agent ever holds standing write access to your cloud. Access is granted for a single approved job, and it expires with that job.

illustration
request › A highly available Postgres database for our EU customers, 100 GB, under $80 a month
Preparing a change for review…
CreatesManaged Postgres · EU
AvailabilityHighly available
Monthly costWithin your ceiling
Replaces or deletesNothing
✓ Measured against your rules
✓ Spending limit respected
● Waiting on a named approver
Nothing has been applied.
A person for anything destructiveNot a setting. Everything else waits for sign-off too, unless your admin chooses otherwise.
No standing accessNothing holds a permanent key to your cloud account.
Confirmed, not assumedWhat's running is checked against what you asked for.
Yours to keepStandard infrastructure code your team can run without us.
The problem

Knowing what you need was never the hard part.

The distance between a clear requirement and safely running infrastructure is made of process — and that is where teams lose weeks they never planned for.

01

The request waits

Every change queues behind someone senior enough to judge it. That person is a bottleneck by design, and goodwill does not make them scale.

02

The review is an investigation

Reviewers reconstruct what was intended from what was written. Most changes are routine, and every one costs the same expensive attention.

03

The evidence arrives late

Who approved a change, and on what basis, gets assembled afterwards from logs, tickets and memory — if it can be assembled at all.

Generative tooling did not close this gap. It made writing the change faster and left everything after it untouched — so the queue moved from authoring to reviewing.

What changes

Four steps, and a person is one of them.

The whole shape of it, from where your team sits.

01 Describe Someone says what they need in plain language. No vendor tooling to learn first, no module to write by hand.
02 Review It comes back as a specific change — what it creates, costs and puts at risk — already measured against your rules.
04 Verify What is running is confirmed against what was asked for — then checked again on a schedule, for as long as it exists.
Who this is for

Three teams, one gap.

The pain reads differently depending on where you sit. Find the version that sounds like your week.

Platform engineering

You are the gate, and the gate doesn't scale.

“Every team wants infrastructure, and every request routes through me personally.”

Your standards get applied whether or not you are in the room, so the golden path stops depending on your availability.

For platform teams →
DevOps & infrastructure

Senior review time is going to routine work.

“My best engineers spend the week reviewing changes that all say roughly the same thing.”

Routine changes arrive already checked, so scrutiny goes where it actually earns its cost.

For infrastructure teams →
Startup engineering

Production-grade practice, without the headcount.

“I can build infrastructure quickly. I can't defend it in a security review.”

Approval, evidence and verification from the first change — years before you could justify hiring for them.

For small teams →
Control

Nothing reaches production unapproved.

Speed comes from the change arriving already checked, not from removing the check — so approving it is a decision, not an investigation.

  • Spending limits and permitted regions are enforced before a human ever looks
  • Your own rules as policy code, or ready-made packs for region residency, required tags, production review, change size and allowed resources. They can only tighten the platform's, never loosen them
  • Anything that would destroy or replace live data requires elevated sign-off
  • A persuasively worded request does not move the boundary: the rules are not something the AI can argue with
  • Automated triggers can propose a change; only a person, or an auto-approval rule your admin wrote, can authorize one
Proof

“It succeeded” is not the same as “it's correct.”

Success tells you the request was accepted, not that your infrastructure now matches what you asked for. Only the second one matters.

  • What is running is checked against what was requested, every time
  • Ongoing checks catch it when reality quietly moves away from your intent
  • When something drifts, Intentric proposes the correction, with its estimated monthly cost from list prices — an estimate, not a bill — and waits for a person
  • Who asked, who approved, what changed and what resulted is recorded, and exported as CSV or JSON or streamed to your webhook
  • When your auditor asks, a control-mapped evidence report over a closed period comes from your own records — a defined subset of SOC 2 Security and ISO 27001 Annex A controls, with the limits stated

Agents propose. Policy and people dispose.

Intentric guiding principle
No lock-in

Everything we build for you is yours.

Intentric produces standard, inspectable OpenTofu/Terraform-compatible infrastructure code — the format your team already reads and reviews. Export it, diff it, commit it, and run it entirely without us.

What you would lose by leaving is the workflow. Not the estate.

Runs on AWS, Azure and Google Cloud

Name a cloud, or leave it open and Intentric picks one on price and on where you already have an account, and shows you the scoring. The same change is written as OpenTofu or as Pulumi, whichever engine your deployment uses.

Not every service is on every cloud. Intentric tells you before it plans.

See the availability matrix →

See it first

Judge it before it touches your cloud.

Three ways in, all held to the same rules: nothing is applied until a person approves it.

Terminal

Stay in your terminal

A command-line tool that runs the same loop as the console, in one file, and drops straight into a CI job.

  • Describe, price, review, apply, check
  • Refuses what the console refuses
  • An apply nobody approved fails the step
Your agent

Let your coding agent draft it

An MCP server lets the agent you already use describe, price and propose a change from your editor.

  • Stops at a priced, checked change
  • No tool in it approves or applies
  • A person decides, or your own auto-approval rule — never for anything destructive

Bring us a change you're nervous about.

Start in the console, or book a demo and we will walk through what the reviewed version of it looks like.