Key Takeaways
By participating along with us in the workshop, you'll learn

Understand the core primitives of an agent: models, tools, history, memory, and orchestration
Implement a tool-calling loop with conversation-style messages instead of relying on a framework
Plug in filesystem tools, 3rd-party tools (web search, code execution, browser handoff), and other SDKs with minimal extra code
Manage context windows with summarization and retrieval so your agent can use local files and search results without blowing the token limit
Design and run evals that cover both single-step decisions and full multi-step runs
Add guardrails and human approvals around risky tools like shell commands or bulk edits
Treat agents as composable building blocks that can be called as tools or wired into other systems

Is This Workshop for Me?
Frontend/full-stack devs who want to build tool-using agents in TS, teams exploring agents for ops/CRM workflows, or any engineers curious about ChatKit-style UIs and chat-embedded forms.

Workshop Details
You’ll build a generic agent that can orchestrate tools on your machine like reading and transforming files, calling 3rd-party tools like web search and code execution, and handing results off to a local browser. The focus is creating an agent with a loop that maintains conversation history, uses tool calling to select tools and arguments, updates messages based on tool results, and decides when to stop. Along the way, you’ll learn how to manage context with summarization and retrieval, layer in evals to catch failures, and add guardrails and human-in-the-loop checks for sensitive actions. By the end, you’ll have an agent you can keep extending with new tools, protocols, and interfaces.