Agent Overview
Agent Overview
Section titled “Agent Overview”Moe’s provides pre-built agent templates that you can deploy out of the box. Each agent is specialized for a specific job and designed to work with other agents through A2A pipelines.
Available Templates
Section titled “Available Templates”| Agent | What It Does | Best For |
|---|---|---|
| Research | Web research, competitive analysis, market intelligence | Due diligence, market reports, fact-checking |
| Support | Email triage, ticket routing, customer replies | Customer support, helpdesk, intake |
| Content | Website copy, blog drafts, content updates | Marketing, documentation, corrections |
| QA | Visual verification, deployment checks, testing | Release validation, monitoring |
| Custom | Whatever you need | Specialized workflows |
How Agents Are Built
Section titled “How Agents Are Built”Every agent is defined by a small set of files in its workspace:
SOUL.md
Section titled “SOUL.md”The agent’s personality and operating instructions. What it does, how it responds, what format it uses for output. Think of it as the agent’s job description.
AGENTS.md
Section titled “AGENTS.md”Technical configuration. Which tools it can access, how it communicates, where it stores its memory.
Skills
Section titled “Skills”Installable capabilities that give the agent specific abilities. A research agent might have a “deep-research-pro” skill that teaches it multi-step research methodology. A QA agent might have a “browser-screenshot” skill.
Memory
Section titled “Memory”Each agent maintains its own working memory in memory/YYYY-MM-DD.md files. This lets agents reference past work and build context over time.
Model Selection
Section titled “Model Selection”Different agents benefit from different AI models. Not every task needs the most powerful (and expensive) model.
| Agent | Recommended Model | Why |
|---|---|---|
| Research | Claude Sonnet / Gemini Pro | Good balance of quality and cost for web research |
| Support | Claude Sonnet | Needs good judgment for customer interactions |
| Content | Claude Sonnet | Writing quality matters |
| QA | Gemini Flash / Claude Haiku | Simple verification tasks, speed over depth |
| Custom (coding) | Claude Opus / Sonnet | Complex reasoning for code generation |
You can configure the model per-agent in your gateway config. Start with the recommended defaults and adjust based on output quality and cost.
Cost Optimization
Section titled “Cost Optimization”For production deployments with many agents running frequently:
- Use cheaper models for high-volume, low-complexity tasks (QA checks, simple routing)
- Reserve premium models for tasks where quality directly impacts customers (support replies, content writing)
- Monitor token usage per agent and adjust as you learn which tasks actually need more reasoning power
Deploying an Agent
Section titled “Deploying an Agent”- Choose a template (or start from scratch)
- Customize the SOUL.md for your specific use case
- Add any skills the agent needs
- Connect it to a Telegram bot
- Add it to your monitoring group
- Test with a simple task
See Quick Start for step-by-step instructions.