Agent
Blueprints
Architecture-first patterns for building LLM workflow and agent systems. Learn to design before you code.
How to navigate this repo
Three entry points, one through line.
Foundations
Core vocabulary, mental models, and how to choose the right pattern for your problem.
Workflows & Patterns
LLM workflows and agent patterns, each documented across overview, design, and implementation tiers.
Interactive Playground
Toggle components on/off in any pattern to see exactly what each piece does.
The catalog at a glance
Patterns shape the control flow. Primitives provide capabilities. Modifiers wrap a pattern with an orthogonal behavior.
Patterns
Cognitive shapes the control flow can take — workflows where the code drives, agents where the LLM drives.
Primitives
Capabilities a pattern draws on — memory across turns, callable tools, agent-discovered skills.
Modifiers
Orthogonal behaviors layered on top of a pattern — human approval gates, audit logging, retries.
The cohorts and entries can grow. taxonomy.yaml and patterns-catalog.yaml at the repo root are the authoritative source — if the site is behind, trust the catalog.
Workflows evolve into agent patterns
Every pattern is documented from high-level overview to implementation detail.
LLM Workflows
The code controls the flow. Developer-defined structure, LLM-filled content.
Evaluator-Optimizer
Generate-evaluate feedback loop that iteratively improves output.
Orchestrator-Worker
LLM decomposes a task and delegates to specialized workers.
Parallel Calls
Concurrent LLM calls on independent inputs, aggregated at the end.
Prompt Chaining
Sequential LLM calls with validation gates between steps.
Agent Patterns
The LLM controls the flow. Developer-provided tools and constraints, LLM-decided actions.
Agentic RAG
RAG where the agent plans retrievals, decomposes queries, routes across sources, reflects on sufficiency, and enforces citation-bound answers.
Event-Driven
Agents triggered by queue or stream events rather than HTTP requests.
Long-Horizon
Multi-session agent tasks that span hours to weeks; checkpoint-and-resume across crashes, deploys, and external waits.
Multi-Agent
Supervisor-worker delegation across multiple autonomous agents.
Plan & Execute
LLM creates a full plan upfront, then executes each step sequentially.
RAG
Retrieval-augmented generation: retrieve relevant context before generating.
ReAct
Reason-act loop: the LLM reasons, calls a tool, observes, and repeats until done.
Reflection
LLM critiques its own output and self-improves through structured feedback.
Routing
Intent classification dispatches inputs to specialized handlers.
Saga
Long-running, multi-step business processes that need compensation when an intermediate step fails.
Ready to design your agent?
Use the decision flowchart to find the right pattern, or explore the evolution explorer to understand the full design space.