Build-An-Agent Workshop
One Wizard. Every Agent SDK.
An educational builder for AI agent CLIs. Answer the wizard once, then download a complete TypeScript project on Claude, OpenAI, GitHub Copilot, or HuggingFace — same questions, same project shape, four runtimes.
Free, open-source, and no lock-in.
Claude Agent SDK
OpenAI Agents SDKPick Your SDK. Keep Your Agent.
Most agent scaffolds lock you into a single vendor. Here, the provider is just one step in the wizard. Your domain, tools, workflows, and permissions carry over — only the SDK underneath changes.
Claude Agent SDK
@anthropic-ai/claude-codeAnthropic's agent framework with native streaming and built-in file and code tools.
- Models
- Sonnet 4.5, Haiku 4.5, Opus 4.1
- Auth
- ANTHROPIC_API_KEY
OpenAI Agents SDK
@openai/agentsThe official OpenAI agent framework, with advanced function calling and a large ecosystem.
- Models
- GPT-5.1, GPT-5 mini, GPT-4.1
- Auth
- OPENAI_API_KEY
GitHub Copilot SDK
@github/copilot-sdkThe same agent runtime as Copilot CLI, with built-in file, search, and shell tools.
- Models
- Auto, Claude Sonnet 4.5, GPT-5, GPT-5.4
- Auth
- Copilot subscription (no API key)
HuggingFace Tiny Agents
@huggingface/tiny-agentsJSON and markdown only. No build step, open-source models, publishable to the Hub.
- Models
- Qwen 3, Llama 3.3, DeepSeek
- Auth
- HF_TOKEN
What stays the same no matter which SDK you pick
Domain, template, tools, and permission level are configured identically across providers.
TypeScript source, slash command workflows, and a permission system you can read and edit.
You own the code. Rerun the wizard on another SDK to see exactly what changes underneath.
Why Build-An-Agent?
An exploration tool for understanding agentic workflows. Not a magic solution.
Not a Silver Bullet
- •Agents can make mistakes and hallucinate. Always verify outputs
- •Complex tasks still need human oversight and judgment
- •This tool helps you learn and experiment, not replace expertise
- •Start small, iterate, and understand what works for your use case
Security-First Mindset
Built for developers, researchers, domain specialists, and curious minds exploring what's possible with agentic workflows.
Enterprise Workflows Meet AI Intelligence
Generated agents include sophisticated slash command workflows for multi-step operations. Built on the Claude Agent SDK, OpenAI Agents SDK, GitHub Copilot SDK, and HuggingFace Tiny Agents with MCP tools, streaming, and configurable security.
Multi-Step Workflows
Domain-specific slash commands like /literature-review, /code-audit, /invoice-batch orchestrate complex multi-step processes. Template variables, retry logic, and error handling built-in.
Configurable Security
Claude Code-style permission system with interactive prompts for file operations, command execution, and network requests. Users approve high-risk actions before they execute.
SDK-Native & Customizable
Download complete TypeScript source code. Full control over prompts, tools, and workflows. Extend with custom business logic from day one.
Zero-Build Tiny Agents
Create lightweight agents with just JSON + markdown. No build step, no dependencies. Run instantly with npx, share to the HuggingFace community, and use powerful open-source models.
See This In Practice
Different people, different workflows. Select your path to see how you might explore agentic tools.
Developer
Software Engineer
Automate code reviews, generate boilerplate, and run systematic audits with natural language commands.
Researcher
Academic / Analyst
Systematic literature reviews, citation management, and evidence synthesis with source tracking.
Business Pro
Domain Specialist
Document processing, report generation, and workflow automation for repetitive tasks.
Hobbyist
Tech Enthusiast
Personal productivity tools, home automation scripts, and learning experiments.
Master the 5 Levers of Control
Claude Code agents are shaped by 5 control mechanisms. Each serves a distinct purpose in the agentic workflow. Click a lever to learn when and how to use it.
Memory
Project context and coding standards
CLAUDE.md
Slash Commands
Reusable task templates
.claude/commands/
Skills
Specialized expertise modules
.claude/skills/
Subagents
Specialized worker processes
.claude/agents/
Hooks
Event-driven automation
.claude/settings.json
Ready to configure these levers for your agent?
Expert Agent Templates
Start with battle-tested templates designed for specific domains and use cases.
Research Ops Agent
Evidence-backed literature reviews, source tracking, and analyst-ready briefs.
Legacy Code Modernization
Analyze legacy codebases, identify technical debt, and create modernization strategies.
Document Processing
Extract, analyze, and transform business documents at scale with automation.
Social Media Manager
Create, plan, and optimize social media content across all major platforms.
Ready to Build Your Workflow-Powered Agent?
Choose your domain, select a template with pre-built workflows, and pick any of the four supported SDKs. Download a TypeScript CLI with slash commands, multi-step orchestration, and configurable security.
Our Vision
Build An Agent Workshop is an educational resource for understanding what goes into AI agents. We believe in learning before automating, open-source community-driven development, and the future of local-first, offline-capable AI tooling.