Standard infrastructure code out. Nothing you can't open.
You are probably the person who gets asked whether this is safe to adopt. Here is the part that matters to you: what Intentric produces is ordinary, readable infrastructure code in the format your team already reviews — not a proprietary artifact that only works while you keep paying us.
You can read what it is about to do.
A platform you cannot inspect is a platform you should not run. Before anything reaches your account you can see the change in full — what it creates, what it costs, what it touches — and stop at that point for any reason or none.
Afterwards, export the code it produced and keep it. It is standard, inspectable and conventional enough that reviewing it is a normal Tuesday for anyone who already reviews infrastructure changes.
# A managed Postgres database, highly available,
# 100 GB, in eu-west-1 — as approved.
resource "aws_db_instance" "app" {
engine = "postgres"
engine_version = "16.4"
instance_class = "db.t4g.medium"
allocated_storage = 100
multi_az = true
storage_encrypted = true
}Illustrative. Ordinary OpenTofu/Terraform-compatible code — no wrapper, no proprietary syntax, nothing that needs Intentric to run. A plan on the Pulumi engine ships as a Pulumi.yaml you can run with the stock Pulumi CLI.
The whole thing works without opening a browser.
The command line runs the same loop as the console: describe what you need, see the change and its cost, apply it once it is approved, and check the verdict afterwards.
- Approval still belongs to a person, and the tool says plainly when it is waiting on one
- A single file with no runtime to install, so it drops into a CI job as easily as your shell
- Exit codes mean what you expect: an apply that was never approved fails the step
- Reads a Terraform directory, says honestly what it can carry over, and refuses a lossy import unless you say so
$ intentric new "a postgres database and an s3 bucket in us-east-1"
created · 2 resources · $86/mo estimated
$ intentric plan
ready for approval · 2 to create, 0 to change, 0 to destroy
$ intentric apply
waiting: this change needs approval from someone else
$ intentric status
applied · verified · every resource checked against what you asked forIllustrative output. The refusal in the third step is the point — one person cannot request and approve the same change.
Your agent can drive it, and still cannot approve anything.
Intentric ships an MCP server: a single file your editor or coding agent runs, signing in as a service account or as the session the command line already holds. It drives the platform up to the point of decision, and no further.
What an agent can do
- Price a plain-language request, record it, and propose a change for an environment
- Read that change with the verdict your rules gave it, and discuss it with the person deciding
- Read the applies that followed
- Read the record: the audit trail, the evidence reports, and the sample an auditor would draw
What it cannot do
- Approve, reject or apply a change
- Link a cloud account, or manage members or service accounts
- Touch your infrastructure's state
There is no such tool, so no setting turns one on. A person decides — or, where you have turned on auto-approval, your own rule does, and never for anything destructive.
You keep working the way you work.
- Bind a definition to a repository on your git host: a change there can be proposed as a plan automatically, and still goes through the same approval
- Separate environments carry separate parameters and separate stakes
- Your own automation gets machine access, held to exactly the same rules as a person
- A command-line tool, for the people who would rather not use a browser, and an MCP server, for the coding agent that would rather not either
What ships with the platform.
- The CLI is one file with a published SHA-256 checksum, released with each version
- The MCP guide, with configurations for Claude Code, Cursor, VS Code with Copilot and Kiro
- The API reference, route by route
- Export a plan's generated code as a self-contained archive and run it with stock OpenTofu, without Intentric
Straight answers, in the order they usually come up.
Can an AI change our infrastructure without a human?
No. Automation — including a correction proposed after drift is detected — can propose a change and nothing more. A named person authorizes every change before it is applied, unless your admin turns on auto-approval for plans that only add new resources, with the environments, resource types and monthly ceiling you choose. Deleting or replacing anything always requires a person, and anything that holds data requires an elevated approver; no configuration changes that.
What happens to our access keys?
Nothing holds a permanent key to your account. Access is granted for approved work and expires when the job ends, successful or not. There is no shared, long-lived service account sitting behind this.
Are we approving a summary or the real thing?
The real thing. Approval attaches to the exact change you reviewed. Alter it and the previous approval does not carry over — you review the new one.
What policy can we enforce?
The platform applies its own rules first: permitted regions, a spend ceiling per environment tier, and a person for anything destructive. You add your own as policy code at organization, project or environment level, or start from five ready-made packs: region residency, required tags, production review, change size and allowed resources. A layer can only tighten the rules beneath it, and every decision records the rules it was made with.
What happens to infrastructure we already have?
It stays where it is. You choose what comes under management and when, starting with whatever is least frightening. Nothing is adopted or altered because it happened to be in the account. Terraform you already wrote can be imported as a proposed definition, with a report of anything that did not carry over; resources you already run can be adopted, under an elevated approval.
Does this fight with our existing tooling?
It sits upstream of it. Because the output is standard infrastructure code, the orchestration and state tooling your team already runs keeps working — Intentric produces the thing those tools were built to run.
Which engine does it use, OpenTofu or Pulumi?
Either. OpenTofu is the default, and a deployment, a request or your build defaults can choose Pulumi instead. A plan on Pulumi ships as a Pulumi.yaml you can run with the stock Pulumi CLI.
What if we leave?
You export the infrastructure code and carry on. You lose the workflow — the requesting, reviewing, approving, verifying and record-keeping. You do not lose your infrastructure, and you do not need our permission to keep running it.
Is a successful apply treated as success?
No, and this is a real distinction. A change that reports success has told you the request was accepted. What is actually running is separately checked against what was asked for, and checked again on a schedule afterwards.
Drive it yourself, against a cloud that isn't yours.
The sandbox runs the entire path against a simulated cloud, and we say so up front. Nothing else about it is: the same product makes every decision, applies every rule and refuses everything it would refuse on your own account.
- A priced quote, a real change to review, a real approval under the same two-person rule, the apply, and the check afterwards
- No signup, no card, no cloud account, nothing to spend
- Try to break the approval rule yourself — better there than wondering about it later
What the sandbox will not do
It will not touch a real cloud account, so it cannot show you a real bill, a real region outage, or how a provider behaves under load. Those are questions for a conversation, and you can book one.
It will also not pretend a step happened. If something is refused, it is refused for the same reason it would be on your account — that is what makes it worth your time.
Ask us the hard version of any of those.
We would rather answer it now than have you find out during a security review.