Your first session

Run one real governed session in about 15 minutes. By the end you will have seen the core loop: session, then approval, then recorded outcome.

Before you start: you need an invitation and a tenant hostname (often slug.preshos.com). No tenant yet means asking your admin, since there is no public signup. Admins: see Day-0.

Step-by-step

1. Sign in at your tenant hostname (2 min)

Open the exact hostname from your invite (for example acme.preshos.com) and sign in.

• The address bar shows your tenant hostname, not someone else’s
• You see your tenant’s organization name after sign-in

My tenants, hostname wins. If you belong to several tenants, the hostname decides which one you are acting in. Same identity, different tenant. When anything looks unfamiliar, check the hostname first.

Sign-in order: invite link, then tenant hostname. If you see a preparing screen, wait and retry (provisioning is still finishing). Otherwise sign in and land on tenant home.

Seeing a /preparing screen is normal on brand-new tenants, since provisioning is still finishing. Wait and retry; do not re-register.

2. Open an Agent Environment (3 min)

From home, open the environment your admin pointed you at. An environment is the isolated workspace your session runs in: which systems it sees (context), what it may attempt (tools), and what needs approval (policy).

• You can see at least one environment
• You know which one matches your task (ask your admin if unsure; wrong environment is the most common beginner mistake)
• You skimmed the environment’s available tools: this list is the source of truth for what the agent can touch

No environments visible? Stop here and ask your admin or the PRESHai team. Do not improvise.

3. Start a session with one small, real goal (2 min)

Start a session and give it one bounded goal. Good first goals are real, verifiable, and low risk.

Avoid: vague goals (“improve things”), multi-system sprawls, or anything irreversible. One goal per session.

• Goal written in one sentence
• Goal fits what the environment’s tools can reach

Good first session	Why
Look up the status of ticket X and summarize it	Real, verifiable, low risk
Draft a reply to ticket Y for my review	Shows draft vs. execute
List devices missing patches in group Z	Scoped read, concrete result

Avoid: vague goals (“improve things”), multi-system sprawls, or anything irreversible. One goal per session.

• Goal written in one sentence
• Goal fits what the environment’s tools can reach

4. Watch it work and meet the approval gate (5 min)

Let the agent act. Reads and drafts flow through; anything above the risk threshold pauses and asks you.

The exchange runs in this order: you give the agent one bounded goal, the agent does reads and drafts directly, the agent sends risky actions to the approval gate, the gate shows you what, on-what, and why, you allow or deny, the gate tells the agent to proceed or stop, and the agent returns a recorded outcome.

When a request arrives, check the what (exact action), the on-what (target, inside session scope), and the why (does it serve your goal). Then allow or deny deliberately. Full decision guide: Approving agent actions.

• You received (or understand why you did not receive) an approval request
• You made at least one deliberate allow or deny decision, or noted why none was needed

5. Read the recorded outcome (3 min)

Every session ends with a recorded outcome: what was attempted, what was approved or denied, and what changed.

• You can state the session’s result in one sentence
• You can find the record again (approvals and denials alike are kept)

What next

• Run your real work the same way: one goal, right environment, deliberate approvals. See Working with sessions.
• Hit something odd? See User troubleshooting.
• Curious how tenants get set up? See Tenant admins.