# Sessions that stay signed in

A login is the slowest thing an agent does and the one most likely to fail. Here is how a saved session skips it.

Ignacy Ruszkowski | Aug 05, 2026 | 4 min read

Source: https://talona.ai/blog/sessions-that-stay-signed-in

Signing in is the part of a task nobody asked for. It is also where most runs are lost: a form that moved, a code sent to a phone, a challenge that only appears the third time an address is seen from a new machine.

A session you saved once does not have to do any of it again. This is what that costs and what it does not cover.

## What is kept

A saved session is the browser state a site uses to recognise a returning visitor, and nothing else:

- **Cookies**, for the origins the run actually touched.
- **Local and session storage**, which is where most single-page apps keep the token the cookie used to hold.
- **The profile’s fingerprint**, so the machine that comes back reads as the machine that signed in.

Nothing else survives the run. The page cache, the downloads, the history and the disk the browser wrote them to go when the session does.

## Using one

Name a session on the way in and the next run that names it starts where the last one stopped:

```python
session = talona.sessions.create(profile="acme-crm")
page = session.page()
page.goto("https://crm.example.com/leads")  # already signed in
```

The first run through does the login by hand, the way a person would. Every run after it is the second paragraph of the task rather than the first.

## What it does not fix

A saved session is not a way around a site that does not want you. Sessions expire, and a site is free to expire yours early — after a password change, a new device rule, or a quiet policy nobody announced.

So the sign-in path has to keep working even when it is almost never taken. Treat it as the fallback it is: a run that finds itself logged out should log in and carry on, not fail and wait for somebody to notice.