Teams & Agents

The Team → Agent Model

Every team you create in Heard gets its own dedicated AI agent. This agent is purpose-built for that team, possessing its own communication style, its own knowledge base, and its own unique set of tools.

When a conversation is analyzed, Heard’s AI determines which team it belongs to (such as Support, Sales, or Retention) and routes it to that particular team’s agent. The agent then reasons about the interaction using that team’s playbook rather than relying on a generic prompt.

---

What Makes Each Agent Unique

Communication Style

Each team’s agent writes differently:

Setting	Example
Tone	Formal (“Dear Mr. Johnson…”), Casual (“Hey Sarah!”), or somewhere in between
Voice	First person (“I’ll look into this”) or Team voice (“We’ll look into this”)
Phrases to use	”We appreciate your patience”
Phrases to avoid	”Unfortunately,” “As per our policy”

The agent applies these style rules whenever it drafts emails, writes action plans, or responds through the OpenClaw.

Knowledge Base (Playbooks)

Each team can upload its own specific documentation containing the rules, policies, and procedures the agent should follow. This can take several forms:

• PDF or DOCX files like your support handbook, refund policy, or sales playbook.
• Google Docs links which are automatically synced and kept up to date.
• Plain text which you can enter directly in the Dashboard.

When the agent needs to make a decision (e.g., “Should I offer a discount to this at-risk customer?”), it searches the playbook for the relevant policy and follows it. It doesn’t hallucinate; instead, it grounds its reasoning securely in your actual documentation.

Assigned Tools

Each team can access different external tools via MCP (Model Context Protocol):

• The Support team might connect to Jira for ticket creation
• The Sales team might connect to HubSpot for pipeline updates
• The Billing team might connect to Stripe for refund processing

Agents can only access the tools assigned to their team, ensuring strict isolation between teams.

---

How It Works in Practice

Admin creates "Retention" team
    ↓
Heard provisions a Retention Agent
    ↓
Admin uploads the retention playbook (PDF)
    ↓
Admin sets tone: "Empathetic, first person, warm"
    ↓
Admin connects Stripe MCP server (for discount processing)
    ↓
Now, when a customer calls threatening to cancel:
  → Heard routes to the Retention Agent
  → Agent reads the playbook for retention offers
  → Agent drafts a personalized email in the configured tone
  → Agent can apply discounts via Stripe if approved

---

Multi-Team Routing

For organizations with multiple teams, Heard’s AI analyzes the conversation intent and routes automatically:

Detected Intent	Routed To
Complaint, bug report, feature request	Support
Pricing inquiry, product interest, partnership	Sales
Cancellation threat, downgrade	Retention
Check-in, renewal, onboarding	Customer Success

The routing is determined entirely by AI analysis, with no manual rules to maintain. If a conversation has mixed intent, the primary intent wins.