Chris Fregly

A runnable, honest tour of the Managed Agents surface: agents, environments, sessions, the event stream, custom tools, memory stores, vaults and MCP, resources, outcomes, multiagent, and the ant CLI. Runs offline by default; --live provisions a real session and tears it down.

A runnable reference for cutting your Claude bill: one honest demo of each platform cost lever (prompt caching, context editing, tool search, programmatic tool calling, adaptive thinking and effort, token counting, Batches). Runs offline by default.

Ground Claude in fresh and document sources with provenance: server-side web search and fetch, document citations that return source spans, and the Files API. Runs offline by default.

The two Claude I/O primitives: structured outputs (a JSON-schema contract on the response) and the memory tool (its six commands, run on a local backend offline). Runs offline by default.