Getting started
Sanaf is one AI employee per workspace. You hire it, give it a job in plain words, connect the apps that job needs, and set what it may do without asking. Everything below takes about twenty minutes and nothing here needs a card.
Last updated 2026-09-04.
Before you start
You need an account on employees.sanafai.com and, if you want the work happening where your team already talks, permission to install an app into your Slack or Microsoft Teams workspace. Everything else can wait.
Signing up gives you $120 of credit and no card is required. That is enough for hundreds of real requests, which is deliberate: you should be able to decide whether this works before you decide whether to pay for it.
The four steps
The order matters. A job with no connected apps can only draft; a connected app with no approval rules is a hire you have to watch. Do all four before you judge it.
- 1
Hire it
Create the workspace and Sanaf exists. There is one employee per workspace, so there is nothing to choose here and nothing to name.
- 2
Give it a job
Pick one of the thirty five written job descriptions or write your own. A job says what it owns, what done looks like, and what it should never touch. Keep the first one small enough that you can tell whether it worked.
- 3
Connect the apps that job needs
Connect only what the job needs. A connection is scoped to the job and can be revoked on its own, so start with one and add the second when the first is clearly working.
- 4
Set what it may do without asking
Rules are set per kind of action, not per app. Start strict: let it draft and read, make it ask before anything that leaves the workspace, and loosen it as it earns the room.
The first week
Keep it in shadow mode for the first few days. It drafts, you review, and you promote it a level at a time. That is not a trial restriction, it is the same thing you would do with a person.
Read the work log rather than watching the thread. It records what was asked, what it did, what it asked you about, and what it decided not to touch. If something looks wrong, correcting it there teaches it, and the correction is versioned.