Mastra Agents Playbook

This playbook contains guides for building six core Mastra agents: Knowledge Retrieval, Frontend Action, Backend Tool Action, Handoff, Relay, and Group Chat.

Knowledge Retrieval Agent

Imagine a minimal agent that can answer questions from your docs. Build a RAG-lite retrieval agent with local markdown files or upgrade later to a vector DB.

Full Knowledge Guide

Build & integrate the knowledge retrieval agent.

---

Frontend Actions Agent

An agent that outputs structured JSON actions (navigate, openModal, toggleTheme) which your frontend dispatcher can safely execute.

Frontend Actions Guide

Emit safe UI actions from the agent.

---

Backend Tools Agent

An agent that calls backend tools or APIs. Example: fetch the weather from an external API, create tickets, update CRM.

Backend Tools Guide

Wire backend tools & external APIs.

---

Human Handoff Agent

An agent that knows its limits. If a human is required, it escalates gracefully by invoking a handoff tool with target + reason.

Human Handoff Agent Guide

Escalate to humans with context.

---

Multi-agent Orchestration (Coordinator) Agent

An agent that queries multiple people, APIs, or sub-agents and returns a consolidated answer. Perfect for “Who is available tomorrow?” type queries.

Multi-agent Orchestration Guide

Fan-out across sources & merge results.

---

Group Chat Agent

An agent that lives in group chats and only responds when explicitly mentioned, e.g. @agent what’s our refund policy?.

Group Chat Agent Guide

Mention-aware agent for group rooms.

---

Next Steps

• Combine these agents in one Mastra project.
• Deploy to your preferred host and connect them into CometChat Dashboard → AI Agents.
• Extend with guardrails, vector retrieval, and advanced workflows.
• Distribute this playbook internally so teams can copy-paste working agent setups.

---