Package Exports

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (agentuse) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

AI Agents as Simple as Markdown

Build and deploy AI agents with just markdown files.
Perfect for automation, CI/CD, and scheduled tasks. No servers, no frameworks, just text files.

Quick Start • Documentation • Examples

🚀 Quick Start

Zero Setup - Try it NOW (10 seconds)

# Run an agent directly from the web - no files, no install!
ANTHROPIC_API_KEY=sk-ant-... npx -y agentuse@latest run https://agentuse.io/hello.agentuse
# Or override the model to gpt-5
OPENAI_API_KEY=sk-... npx -y agentuse@latest run https://agentuse.io/hello.agentuse -m openai:gpt-5

Create Your Own (30 seconds)

Step 1: Create `daily-reporter.agentuse`: ```markdown

model: openai:gpt-5

Generate a daily motivation quote with an interesting fact about technology. Format as JSON with 'quote' and 'fact' fields.


**Step 2:** Run it:
```bash
# Try without installing (needs OPENAI_API_KEY for this example)
OPENAI_API_KEY=sk-... npx -y agentuse@latest run daily-reporter.agentuse

# Or install globally for production use
npm install -g agentuse
agentuse run daily-reporter.agentuse

# Schedule it (cron, CI/CD, serverless)
0 9 * * * agentuse run daily-reporter.agentuse >> daily-quotes.json

That's it! Your AI agent runs anywhere - CI/CD pipelines, cron jobs, webhooks, or serverless functions.

🎯 Real-World Automation Examples

Daily Metrics Reporter

---
model: openai:gpt-5
description: Daily sales metrics reporter - runs daily at 9am via cron
mcpServers:
  postgres:
    command: uv
    args: ["run", "postgres-mcp", "--access-mode=restricted"]
    requiredEnvVars:
      - DATABASE_URI
---

Query sales_metrics table for yesterday's data.
Generate executive summary with trends and alerts.
Format as markdown report.

---
model: openai:gpt-5
description: SEO performance analyzer - runs weekly via GitHub Actions
mcpServers:
  dataforseo:
    command: "npx"
    args: ["-y", "dataforseo-mcp-server"]
    requiredEnvVars:
      - DATAFORSEO_USERNAME
      - DATAFORSEO_PASSWORD
---

Analyze SEO performance for https://blog.example.com/blog-post
Compare rankings with top 3 competitors in our niche.
Generate keyword opportunities and content gap analysis.
Output recommendations as JSON for our CMS.

---
model: openai:gpt-5
description: Social media automation bot - runs every 6 hours via cron
mcpServers:
  twitter:
    command: npx
    args: ["-y", "@enescinar/twitter-mcp"]
    requiredEnvVars:
      - API_KEY
      - API_SECRET_KEY
      - ACCESS_TOKEN
      - ACCESS_TOKEN_SECRET
  exa:
    command: npx
    args: ["-y", "exa-mcp-server", "--tools=web_search_exa"]
    requiredEnvVars:
      - EXA_API_KEY
    disallowedTools:
      - deep_researcher_*
---

Search for trending tech topics using Exa.
Generate 5 engaging posts based on current trends.
Choose the best one and post to X.

Why AgentUse? The philosophy behind the project...

The Problem

Current agent frameworks force an impossible choice. Visual workflow tools give you drag-and-drop simplicity but create version control nightmares and vendor lock-in. Traditional code frameworks offer power and flexibility but require hundreds of lines of boilerplate just to say "hello world."

The Insight

Claude Code proved that markdown configuration is incredibly powerful for AI interactions. But Claude Code is an interactive CLI tool, not a framework for building deployable agents. What if we took that brilliant markdown-first philosophy and built a proper agent development framework around it?

The Solution

AgentUse makes your agents just markdown files. Not configuration files that generate code. Not visual flows that compile to JSON. The markdown IS the agent. This means:

Version control just works - diff, review, and merge agents like any other code
Share agents with a URL - as easy as sharing a gist
Zero learning curve - if you can write a README, you can build an agent
Production-ready - built-in retries, streaming, error recovery, and MCP support

AgentUse is Infrastructure-as-Code philosophy applied to AI agents. Your agents are text files that can be versioned, reviewed, tested, and deployed like any other code artifact.

✨ Features

🚀 Performance	🔧 Developer Experience	🔌 Integrations
• Sub-second startup • Minimal dependencies • Smart context management • Automatic retries	• Plain markdown files • Zero boilerplate • Git-friendly • URL-shareable agents	• MCP servers • Multiple AI providers (Anthropic, OpenAI, OpenRouter) • Plugin system • Sub-agent composition

📦 Installation & Setup

Quick Try (No Install)

# Run any agent without installing
npx -y agentuse@latest run your-agent.agentuse

Production Install

npm install -g agentuse
# or: pnpm add -g agentuse

Authentication

# Interactive login (recommended)
agentuse auth login

# Or use environment variables
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export OPENROUTER_API_KEY="sk-or-..."

📚 Full installation guide → 📘 Authentication docs →

📚 Documentation

🚀 Getting Started	📖 Guides	📘 API Reference	💡 Templates
5-minute tutorial	Learn concepts	Complete reference	Example agents

📋 Core Concepts

Agents are markdown files with YAML frontmatter for configuration and markdown content for instructions:

---
model: anthropic:claude-sonnet-4-0  # Required: AI model
mcpServers: {...}                   # Optional: MCP tools
subagents: [...]                     # Optional: sub-agents
---

Your agent instructions in markdown...

📚 Agent syntax guide → 📘 Model configuration → 🔧 MCP servers → 🤖 Sub-agents →

🤝 Contributing

We welcome contributions! Here's how to get started:

📖 Read our Contributing Guide
🐛 Report bugs via GitHub Issues
💡 Share ideas in Discussions
🔧 Submit PRs with improvements
⭐ Star the repo to show support!

📜 License

Apache License 2.0 - see LICENSE file for details.

Made with ❤️ by the AgentUse community
GitHub • Documentation • Website

agentuse