Pilae Agent: the AI operator for your open-source apps

It watches every app you run, plans each update, migration and fix, waits for your approval, applies the change in your window and records what it did.

Start with one workloadBook a 30-minute briefing

Approval
Every change, or only outside the rules you set
Runs
Inside your perimeter, next to the apps it operates
Record
Plan, approval, result and rollback for each change
Escalation
Pilae engineers, around the clock

An operator that never skips the boring steps

Most outages in a self-hosted estate are a skipped step: the backup nobody checked before the upgrade, the release note nobody read. The agent does every step, every time, and shows its work.

Updates, planned

It reads the upstream release, checks it against your configuration, and writes a plan with the rollback before it asks.

Security fixes, first

Advisories for the apps you run are matched against your versions. A critical fix is planned and on your desk within hours.

You approve

Nothing changes without a yes, from you or from a rule you wrote. Approve from the console, by email or in chat.

Backup before every change

A verified backup is a precondition of every plan. If the backup fails, the change does not start.

Every action on record

The plan, the approval, each command and its result, kept in the console and exportable to your SIEM.

Humans on call

When a change does not go to plan, the agent rolls back and a Pilae engineer takes over.

An agent that works like your best operator

A good operator reads the release notes, takes a backup, checks it, applies the change in the quiet hour and writes down what they did. The Pilae Agent does the same, for every app in your estate, every time, and it never gets bored of the steps that matter most.

The difference from a script is judgement: the agent reads the release, compares it with your configuration and your history, and writes a plan a person can read. The difference from a person is that it never skips the backup check at two in the morning.

Built for estates that have to answer for every change

Every plan, approval and command is recorded. Your security team gets an audit trail per app. Your auditors get a log that shows who approved what and when. Your engineers get their evenings back.

How a change runs

  1. Discover

    A new release, an advisory, a failing probe or a request from your team opens a task.

  2. Plan

    The agent writes the steps, the expected downtime, the backup it will take and the rollback.

  3. Approve

    You read the plan and approve it, or a rule you set approves it for you.

  4. Apply

    In your maintenance window, after a verified backup, with health checks at each step.

  5. Verify and record

    Probes confirm the app answers. The whole run is written to the audit log.

What your contract includes

Approval rules
Set per app and per change type. The default is that everything asks.
Maintenance windows
Agreed with you, per app. Emergency security changes are proposed, never forced.
Audit log
Retained for the length of the contract and exported on request.
Escalation
A named Pilae engineer on every failed change, with response times per plan.

The apps behind it

Questions

Does the Pilae Agent change anything without approval?

No. By default every change waits for a person to approve it. You can write rules that approve routine changes, such as patch releases inside a window, and the agent still records each one.

Where does the Pilae Agent run?

Next to the apps it operates: on your premises when your apps run there, or on your dedicated machines in Pilae Cloud. Your data does not leave that environment for the agent to work.

Which AI model does the agent use?

The planning model is configurable. Regulated customers can run it on a model hosted inside their own perimeter, so no prompt or log line leaves it.

What happens when a change fails?

The agent stops, rolls back to the backup it took before starting, and hands the task to a Pilae engineer, who contacts you with what happened and what comes next.

Related

See the agent plan a change on your stack.

Thirty minutes with an engineer: we connect to a test app and let the agent write its first plan in front of you.