1. Content Creator Flow ("Inkwell")
Topic → Research crew (Serper + Scrape) → Outline crew (hierarchical with manager) → Draft crew (knowledge from PDFs) → @human_feedback review → MCP CMS publish. Persisted with @persist, streamed to a UI, traced to Langfuse.
2. Recruitment Flow
JD parse → candidate retrieval (CSV/JSON knowledge) → match-score per candidate (kickoff_for_each_async()) → email drafts → human approval gate.
3. Trip Planner ("Surprise Trip")
User profile + budget → destination shortlist crew (Serper, web scraping) → itinerary crew (per-day planner) → calendar tool → notify user.
4. Lead Score Flow
CRM webhook trigger → enrichment crew (web + LinkedIn) → scoring agent (reasoning=True) → router by score → assign to rep or nurture queue.
5. Meeting Assistant
Transcript → summary agent → action-item extractor → Slack/Email tool → calendar follow-ups. All as LiteAgent instances inside a Flow for lean perf.
6. Code Reviewer Crew
PR diff (GithubSearchTool) → review agent (reads CONTRIBUTING.md via knowledge) → security checker agent → bot comment via webhook.
7. Email Auto-Responder
Inbound → classifier → response drafter → @human_feedback for sensitive categories, auto-send for routine ones. Webhook trigger; HMAC-signed inbound.
8. Financial Modeling Crew
Statements (PDFKnowledgeSource + CrewDoclingSource for tables) → analysis agent with CodeInterpreterTool → narrative writer → output to Excel.
9. Support Triage
Ticket in → classifier (LLMGuardrail enforced categories) → router → specialist agents per category → human escalation for P0.
10. Sales SDR
Account list → research crew (Serper + Scrape) → personalized outreach drafter → CRM logging tool → human review for first batch, auto-send after.
Notes
Patterns are templates, not drop-in products
Swap domain data, tighten guardrails, and re-benchmark costs for your traffic. Cookbook flows assume reasonable rate limits and benign tool surfaces.
Combine Flow and Crew deliberately
Mixing orchestrators adds state to reason about. Keep crew boundaries aligned with business roles and use Flow for branching, retries, and external I/O.
Secrets never belong in copied snippets
Parameterize endpoints and keys through environment variables and secret managers. Forked examples often leave placeholder hosts that leak in screenshots.
Add integration tests per pattern
A pattern that demos well on toy data may fail on messy PDFs or slow CRM APIs. Capture one deterministic fixture per pattern you ship to customers.
CrewAI cookbook FAQ
What is the CrewAI cookbook on DevShelfHub?
It is a pattern library of complete crews and flows you can adapt: content factories, recruiting, outbound sales, support triage, and more, each described with realistic orchestration choices.
Can I copy a cookbook pattern into production?
Treat each pattern as a starting skeleton: swap tools, tighten prompts, add guardrails, and wire your own auth, storage, and tracing before exposing anything to users.
Do cookbook examples use Flows or Crews?
Many patterns combine both: Crews handle agent-heavy work while Flows manage branching, human approvals, and persistence. The pages call out where each boundary sits.
Which pattern should I start with?
If you are new, begin with a two-agent sequential crew like research to writer, then graduate to Flow-backed patterns once you need branching or external callbacks.
Where do I learn the primitives behind these patterns?
Read the core concepts, defining agents, and creating tasks lessons in this series, then open the CrewAI API reference index linked from each tutorial page.
See also: DevShelfHub's CrewAI tool review for a product-level comparison, pricing notes, and links back into this tutorial series.