How Aegara works

Your AI systems do their work. Aegara records what each action touched, checks the values inside it where those values already are, and keeps the record of both.

What happens to one action

From the moment your AI acts to the moment you see it, the action is checked on your side of the line, and only its record crosses.

Step one

Your AI takes an action

It reads an email, drafts a reply, calls an API or writes to a database, as it normally would.

Step two

Aegara’s code checks it, inside your environment

The recorder in your app, or the gateway in front of your assistants, records what the action touched and checks the values in it, on your own machines.

Step three

Only the record crosses the line

The record of what happened, and the verdicts on its values, are sent to Aegara. The values themselves stay where they are.

Step four

You see it

Each run appears in the Trace Portal as one trace. When a check catches something, Aegara flags the run and marks the step where it happened.

What leaves your environment

The values inside your AI's tool calls are checked where they already are, on your own machines. Only the verdict is sent.

Aegara's rules run here, on your machine.The values never move.

The verdict, and nothing else. Never the number, a piece of it, or a hash of it.

Your environmentAegarathe values in your AI’s tool callscrm.contactsdocs.policiesfinance.forecastbilling_refthe boundarybilling_ref contained a card numberhigh confidence, payment data Your environmentthe values in your AI's tool callscrm.contactsdocs.policiesfinance.forecastbilling_refthe boundaryAegarabilling_ref contained a card numberhigh confidence, payment data

Think of a smoke alarm. The manufacturer decided what smoke looks like. The alarm sits in your house, sniffs your air, and if it finds smoke it sends one signal: smoke, kitchen. The manufacturer never smelled your air. They wrote the rule; your device did the smelling.

What one trace looks like

Someone asks their assistant to summarize an email thread and draft a reply. This is how Aegara records it.

TransformCallTransform
  1. Fields read
    • Subject
    • Sender
    • Date
    • Body
  2. CallAsked a model running on this machine (llama3.1)
  3. TransformSummarized the thread
  4. CallAsked a model hosted by OpenAI (gpt-4o)
  5. TransformDrafted a reply
  6. WriteSaved a reply as a draft, in the mailbox's drafts

A log records what an action did. After an incident, the question is almost always what else it touched. Aegara records each step of the run, and flags the run when one crosses a line.

What you get

Sensitive values caught where they are

The values your AI reads, sends and hands to a model are checked for sensitive data inside your own environment.

Your whole team on one record

Each person signs in with their own account. Owners manage the team, and every member sees the record.

Proof of who did what

Aegara keeps a trail of who did what inside your organization, and you can download it on every plan.

A statement every month

Each month, Aegara says how many runs it checked and which it flagged, ready for your auditor.

How Aegara installs

Aegara goes in beside what you already run, and nothing in your code changes.

In a Node or Python app

One setting, or starting the app with the aegara run command, loads the recorder before your code’s first line.

On AWS Lambda

Add the Aegara layer to the function and set two values on it. The next invocation is recorded.

In front of your staff’s AI assistants

Run the gateway in your own network and point your assistants at it. For a desktop assistant, one command puts the gateway in front of its local tools.

The recorder reads inside your app, in memory, the way your app does. No content it reads is written to disk, and only names, counts and verdicts leave.

Accounts are by invitation. Request access, and someone at Aegara will reply within one business day.

Request access