Your data stays local. Inference does not.

The short version

• Your data stays local

  History, bookmarks, chats, scheduled tasks, and run logs are files on your machine. We run no server that receives them, so there is nothing of yours for us to hand over, sell, or lose.

• Inference does not, unless you run a local model

  A model request leaves your machine and goes to the provider serving the model you picked. On the free plan it goes to a model server you run yourself, and then it does not leave at all.

• Nothing you send trains a model

  Aera's OpenRouter account excludes endpoints that train on request data, so a request never routes to one. That is a routing rule rather than a promise from us. It covers the models we host, not a local endpoint you run.

• Your own browser profile, not a bot harness

  Aera drives the Chromium profile on your machine, in the sessions you are already signed in to. There is no headless fingerprint and no handoff of your passwords to a cloud runner.

• Local-first, not offline

  Your account, a session check at launch, update checks, and billing amounts do leave your machine. The full list is further down this page.

What stays on your machine

These are files in your Aera profile folder. We hold no copy and run no server that receives them. Settings > Privacy deletes all of it at once, and uninstalling takes it with you.

Browsing history

Every page you visit, in the same local database Chromium uses.

Bookmarks and saved passwords

Stored on your device by Chromium, the same way Chrome stores them.

Chats with the agent

Every conversation, including the page text the agent read in order to answer, in a local database.

Scheduled tasks and run history

The task, its schedule, and the step-by-step log of every run it made.

Skills and files

Skills you write, and files the agent produces, sit on your disk. A skill stays local until you choose to publish it to the marketplace.

Where inference goes

The agent cannot read a page without sending it to a model. No local-first design changes that, so here is exactly where it goes and what is in it.

On a paid plan

The request goes from your machine to OpenRouter, which routes it to the provider serving the model you picked. For frontier models that is the model's own company, such as Anthropic, OpenAI, Google, or xAI. For open-weight models it can be another company that hosts them. The request does not pass through an Aera server.

On the free plan

The request goes to the model server you entered in Settings > Models, which is an address on your own machine by default. Nothing goes to OpenRouter and nothing goes to us. Chat titles, suggestions, and history compaction run on that same local model.

What a request contains

Your instruction, the text of the page, and the earlier turns of that conversation. Also a screenshot, when the model has to see the layout, and any file you attached.

What we never receive

We run no models and keep no copy of a request. Aera's own servers exist for accounts, billing, updates, and the marketplace.

What is not local yet

This is the section that usually gets left out. Aera is local-first, not offline. Everything below leaves your machine on every plan, including the free one.

Your account

Aera requires a sign-in, so your name, your email, and your plan live on our servers. That holds on the free plan too: the agent will not run while you are signed out.

A session check at launch

When Aera starts it asks our server whether your session is still valid. The request carries your Aera version, your platform, a device fingerprint, and the identifier the installer brought from our download page, and we store it with your IP address and user agent. That is how we match a download to the install that followed it, and how we keep one person from collecting referral rewards or free credits as though they were several.

One install event

The first time a new install launches, Aera sends a single event that says so, with the same fingerprint. It is how we tell a download apart from an install that actually ran. It carries no URL and no page content.

Update checks

Aera asks our update endpoint for the current version, sending its own version, platform, and release channel. No account and no cookie ride along.

What you spent

For billing we reconcile the dollar amount of model usage on your key. We receive amounts. Not prompts, not page content, not the model's answers.

The marketplace

Installing a skill increments a public counter for that skill. Publishing one uploads the skill you wrote. Nothing else about your browser goes with it.

We do not ship a model

Running a model locally means installing a server such as Ollama yourself. Aera bundles no model and runs no inference on its own, so we are not claiming an on-device model we have not shipped.

The risks we are not going to hide

An agent that can act in your browser is useful for the same reason it is dangerous. These are the parts of the design you should weigh before you point it at something that matters.

Prompt injection is real

A page can carry text written to redirect any agent that reads it. Aera reads pages. Treat a run on an untrusted page the way you would treat running a script from that page, and read the run log afterwards.

The agent works inside your signed-in sessions

That is the whole point of it, and it is also the exposure: what you can do while signed in, the agent can be told to do. Aera ships no payment feature and no autonomous checkout. It has no card and no wallet. What it has is your browser, so scope what you ask it for.

Local connectors run with your permissions

An MCP connector you install runs as an ordinary process under your own user account, with the access that gives it. Install ones you would trust with that.

No compliance certification

We hold no SOC 2 and no ISO 27001. If your review needs one, we are not there yet, and we would rather say so than let you find out during procurement.