Policy Packs

Policy Packs are sets of rules that agents must follow. They're defined at the identity level and can be attached to agents to ensure consistent behavior and compliance.

What are Policy Packs?

Policy Packs are:

  • Sets of rules: Brand style, safety, compliance, operational
  • Identity-level: Defined once for the entire identity
  • Attachable: Agents can use one or more policy packs
  • Enforceable: Rules can block, require approval, or warn

Think of policy packs as "guardrails" that keep agents on track.

How Policy Packs Work

1. Defined at Identity Level

Policy packs are created in Settings → Policy Packs:

  • One definition for the entire identity
  • Available to all agents in that identity
  • Centralized management of policies

2. Attached to Agents

Agents select which policy packs to use:

  • Choose packs: Which policies apply to this agent
  • Enforcement level: Block, require approval, or warn
  • Precedence: How to resolve conflicts

3. Enforced During Operation

When agents act:

  • Policies are checked: Rules are evaluated
  • Action is taken: Block, escalate, or allow
  • Violations are logged: For audit and review

Creating Policy Packs

Step 1: Go to Settings

  1. Go to Settings → Policy Packs
  2. Click "Add Policy Pack"
  3. Fill in policy pack details

Step 2: Define Policy Pack

Name

  • Display name for the policy pack
  • Example: "Brand Style", "Safety Core", "Ops Caps"

Description

  • What this policy pack covers
  • Example: "Brand voice, tone, and style guidelines"

Category

  • Type of policy pack
  • Examples: Brand, Safety, Compliance, Operational

Step 3: Define Rules

Policy packs can include rules for:

Brand & Style

  • Tone bands: Professional, friendly, formal, casual
  • Signatures: Required email signatures
  • Formatting: How to format messages
  • Language: Preferred terminology

Safety & Compliance

  • Prohibited topics: What agents can't discuss
  • PII handling: How to handle personal information
  • Legal disclaimers: Required disclaimers
  • Compliance rules: Industry-specific rules

Operational Caps

  • Spending limits: Maximum amounts
  • Outbound rate limits: Messages per hour/day
  • Attachment rules: File size, type restrictions
  • API call limits: Rate limits for integrations

Data Governance

  • Retention: How long to keep data
  • Export rules: When data can be exported
  • Third-party sharing: Rules for sharing data
  • Privacy: Data privacy requirements

Escalation Policy

  • Who approves what: Approval requirements
  • When to escalate: Escalation triggers
  • Timeouts: How long to wait
  • Fallbacks: What to do if no response

Step 4: Set Enforcement

Enforcement Level

  • Block: Prevent action if policy violated
  • Require Approval: Escalate for review
  • Warn: Log violation but allow action

Scope

  • All agents: Apply to all agents by default
  • Specific agents: Apply to selected agents only
  • Agent-specific: Override per agent

Step 5: Save Policy Pack

  1. Review configuration
  2. Click "Save"
  3. Policy pack is now available to agents

Using Policy Packs in Agents

Step 1: Attach Policy Packs

  1. Go to Agent Config → Policies
  2. Select policy packs to attach
  3. Choose enforcement level
  4. Set precedence rules

Step 2: Configure Enforcement

Attached Packs

  • Check which policy packs apply
  • Agent must follow all attached packs
  • Example: Brand Style + Safety Core

Enforcement Level

  • Block: Prevent action if violated
  • Require Approval: Escalate for review
  • Warn: Log but allow

Precedence

  • Identity overrides: Identity policies win
  • Most restrictive wins: Apply strictest rule
  • Choose based on your needs

Step 3: View Effective Policy

Effective Policy View

  • See what rules actually apply
  • Combines identity and agent policies
  • Shows enforcement level
  • Helps understand what agent must follow

Policy Pack Examples

Brand Style Pack

Purpose: Ensure consistent brand voice

Rules:

  • Tone: Professional, friendly
  • Signature: Required company signature
  • Formatting: Use company templates
  • Language: Use approved terminology

Enforcement: Block (must follow brand guidelines)

Safety Core Pack

Purpose: Prevent harmful or inappropriate content

Rules:

  • Prohibited topics: Legal advice, medical advice
  • PII handling: Never share personal information
  • Disclaimers: Required legal disclaimers
  • Safety: No harmful or offensive content

Enforcement: Block (safety is non-negotiable)

Ops Caps Pack

Purpose: Prevent overspending and abuse

Rules:

  • Spending: Max $100/day per agent
  • Rate limits: Max 100 emails/day
  • Attachments: Max 10MB, no executables
  • API calls: Max 1000/day

Enforcement: Require Approval (escalate if exceeded)

Policy Precedence

When multiple policies apply:

Identity Overrides

  • Identity-level policies override agent policies
  • Ensures identity-wide consistency
  • Example: Identity says "professional tone" → Agent can't use "casual"

Most Restrictive Wins

  • Apply the strictest rule
  • Ensures maximum safety
  • Example: One policy blocks, one warns → Block wins

Agent Overrides

  • Agent-specific policies can override identity (if allowed)
  • Gives flexibility for special cases
  • Example: One agent can use different tone for specific use case

Best Practices

Start with Core Policies

  • Create essential policy packs first
  • Brand style, safety, basic operational caps
  • Add more as needed
  • Don't over-complicate initially

Be Specific

  • Clear, specific rules work better
  • "No refunds over $100" not "be careful with refunds"
  • Easier for agents to follow
  • Easier to enforce

Review Regularly

  • Review policy violations
  • Update policies based on issues
  • Remove unnecessary policies
  • Keep policies current

Document Policies

  • Clear descriptions of what each policy does
  • Examples of allowed/blocked actions
  • Help team members understand
  • Useful for audits

Troubleshooting

Policies not being enforced?

  • Check agent has policy pack attached
  • Verify enforcement level is set
  • Review policy pack configuration
  • Check agent logs for violations

Too many violations?

  • Review policy rules (may be too strict)
  • Adjust enforcement level (warn instead of block)
  • Update policies to match actual needs
  • Provide feedback to agents

Policies conflicting?

  • Review precedence settings
  • Choose identity overrides or most restrictive
  • Resolve conflicts explicitly
  • Document resolution approach

Next Steps

Version: 1.0
Last Updated: November 2025