Curriculum
Five in-depth articles that explain the core concepts — from building your first ReAct loop to running comprehensive agent evaluations.
Reason · Act · Observe · Repeat
Bounded ReAct and reflection loops that stop cleanly.
Schema · Validate · Retry · Normalize
Typed schemas, resilient wrappers, and safe tool calls.
Window · Summarize · Retrieve · Evict
Context budgets, durable state, and retrieval handoffs.
Chunk · Embed · Rerank · Ground
Retrieval, reranking, grounding, and evidence shaping.
Measure · Judge · Trace · Improve
Deterministic checks before judgment-layer evaluation.
Validate · Approve · Restrict · Recover
Permissions, approval gates, and safe failure modes for tool-using agents.
Trace · Log · Replay · Improve
Traces, logs, replay, and metrics for understanding agent behavior.
Connect · Expose · Secure · Use
Building MCP servers and connecting agents to external systems.