---
title: "Give Pioneer AI instructions"
description: "Tell Pioneer AI how your team works, per organization and per client, and add rules the AI safety review checks before it approves an action."
url: "https://support.outpostplatform.com/pioneer-ai/instructions/"
product: "depot"
type: "how-to"
status: "draft"
ai_summary: "Settings > AI has Instructions for Pioneer AI (4,000 characters) and a Safety review policy (2,000). On the MSP plan each client's AI card adds its own. Instructions rank below Pioneer AI's own rules. The policy can only make automatic approval stricter, and the review sees the current time in the technician's time zone. Needs ai:manage."
availability:
  plans: ["home-lab", "essentials", "enterprise", "msp"]
  permission: "ai:manage"
  preview: false
source: "https://gitlab.com/outpostplatform/docs/-/edit/main/src/content/docs/pioneer-ai/instructions.md"
license: "CC BY 4.0"
---

# Give Pioneer AI instructions

Instructions tell Pioneer AI how your team works: naming conventions,
preferred tools, who to escalate to. A safety review policy adds rules that
the [automatic approval](/pioneer-ai/automatic-approval/) review checks before
it approves an action, such as "never restart servers during business hours".
This page shows how to set both, for your organization and, on the MSP plan,
for one client.

## Before you begin

- You need `ai:manage`, which the owner holds.
- Instructions guide Pioneer AI. They are lower priority than its own rules,
  so they cannot give it permissions, lift an approval, or allow an action
  that is never allowed.
- A safety review policy matters only where automatic approval is on. Without
  it, a technician approves every change anyway.

## Set organization instructions and policy

1. Go to **Settings** > **AI**.
2. In **Instructions for Pioneer AI**, enter how your team works, up to 4,000
   characters.
3. In **Safety review policy**, enter the rules the review must check, up to
   2,000 characters.
4. Select **Save**.

Instructions apply from the next message sent in any conversation.

## Set instructions and policy for one client

On the MSP plan, each client can add its own:

1. Go to **Clients** and open the client.
2. On the **AI** card, enter **Instructions for this client**, up to 4,000
   characters. They are added after your organization instructions in
   conversations about this client.
3. Enter **Safety review policy for this client**, up to 2,000 characters. The
   review checks it together with the organization policy for actions on this
   client.
4. Select **Save**.

## How the safety review uses a policy

When automatic approval would approve an action, and a safety review policy
or a [team memory](/pioneer-ai/memories/#memories-and-the-safety-review)
applies to it, the review checks the action a second time against those
rules. The action is approved only if both checks allow it. A policy can only
make the review stricter: if a rule does not clearly apply, it does not block
the action, and no rule can approve something the first check refused. When
the policy blocks an action, the approval card shows the review's reason and
waits for the technician.

The review states the current time in the technician's time zone (their time
zone preference in the console, otherwise their browser's), so rules about
business hours or maintenance windows work in local time. It uses that clock,
never a time written in the conversation.

Instructions never reach the safety review, and neither do personal memories.

## Verify

- **Instructions.** Start a new conversation and ask something they cover,
  such as how to name a new device group.
- **Policy.** With automatic approval on, ask for an action your policy
  forbids. The approval card waits for you and shows the review's reason.

Every change to these settings is recorded in the
[audit log](/depot/audit-log/).

## Next steps

- [Turn on automatic approval](/pioneer-ai/automatic-approval/)
- [Manage Pioneer AI memories](/pioneer-ai/memories/)