Tactical, step-by-step, no fluff.
The tracks teach the method. Playbooks are the specific builds: cut your API bill, run open models through Claude Code, keep long agents from drifting. Each one is the actual meat, distilled from the best work out there.
Build in Claude Code without the surprise bill
Five small habits that keep your AI spend low, shown as the exact things you type. No code, no jargon.
Read it→Run Claude Code on open-source models via OpenRouter
Point the Claude Code harness at GLM and other open models through OpenRouter, step by step, and know when it is worth it.
Read it→Run Claude Code directly on GLM, Kimi, or DeepSeek
Skip the middleman. Point Claude Code straight at a provider's own endpoint for the lowest price, often a flat coding plan, with the trade-offs mapped.
Read it→Route each task to the cheapest model that can do it
Send the grunt work to a cheap open model and keep the hard thinking on Claude, automatically, with one small local router.
Read it→Keep long agent runs from blowing the context window
Compaction, context editing, and memory: when to reach for each so a long-running agent stays coherent and cheap.
Preview and join the list→