Chapter 10 · vs.
Cursor helps you write code.
Luna helps you run engineering workflows.
Different problem. Different shape. Luna doesn't compete with the editor; it composes with it. Compare on substance, not adjectives.
The category map.
Most tools fit cleanly into one of five categories. Luna sits in a sixth: AI shell language. Adjacent to all of them, replacement for none.
| Product | Mental model | Surface | Composition | Open source | Use it for |
|---|---|---|---|---|---|
| Cursor | AI pair programmer | IDE | implicit (chat) | no | Write code, fast. |
| Claude Code | AI terminal operator | CLI + IDE | slash commands | no (CLI) | Edit a repo conversationally. |
| LangGraph | Agent workflow engine | Python lib | graph DSL | yes | Build a custom agent app. |
| CrewAI | Multi-agent orchestration | Python lib | role/task config | yes | Coordinate roles + tasks. |
| n8n / Zapier | Visual automation | Web UI | drag-drop graph | n8n yes, Zapier no | Connect SaaS apps, no code. |
| Willow | AI agent governance / IAM | Enterprise SaaS | policy dashboard | no (enterprise SaaS) | Govern + audit any agent your org runs at scale. |
| Luna Pipes | AI shell language | CLI + Claude Code + CI | pipe operator (>>) | yes (MIT) | Run engineering workflows, one line at a time. |
Specific comparisons.
vs. Cursor
Cursor is a code editor with AI inside it. Luna is a shell for composing AI workflows. They share zero overlap by design. Use Cursor to write; pipe through Luna to review, test, audit, ship. You can keep both.
vs. Claude Code
Luna ships as a Claude Code plugin. We don't compete; we extend. Claude Code is the runtime; Luna is the verb set. The pipe operator is what's new.
vs. LangGraph / CrewAI
Frameworks ask you to write Python that constructs a graph. Luna asks you to write one line in your shell. If you're building a bespoke agent product, use LangGraph. If you're trying to ship your own features faster, use Luna.
vs. n8n / Zapier
Visual graphs are great for non-coders gluing SaaS. Luna is for engineers who already live in the shell. The pipe expression is version-controllable, diffable, greppable — the graph is not.
vs. Willow
Adjacent, not competitive. Willow is an enterprise IAM platform for AI agents — identity per agent, 1000+ tool connectors, audit forwarded to SIEM, OWASP LLM06 framing. It lives above agents that already exist and govern them at org scale.
Luna lives below. It scaffolds the agent (typed contract, MCP tool list, OPA policy bundle, six-table audit schema, approval inbox, golden-eval harness) — the artefact Willow then governs. The two compose naturally:
/ll-agent-build + /ll-export-governance target=willow
emits an agent manifest Willow can ingest, with identity, scoped
tool permissions, audit endpoint, and the OPA bundle attached.
No double-spec. The same scaffold ships to Cloudflare and to
Willow's policy graph from one source of truth.
Buying Luna does not displace Willow. Buying Willow does not displace Luna. Different layers of the same stack.
When Luna is the wrong choice.
- You're a non-coder. Use n8n or Zapier.
- You're building a customer-facing agent product. Use LangGraph or the Anthropic SDK directly.
- You don't use Claude Code or a Unix-style terminal. The pipe metaphor is the value; without it you're paying tax for no benefit.
Where to start.
- Frontispiece — the recipes, the starter ten.
- Showcase — the four hero commands.
- Install — 30 seconds, idempotent.
- Playground — try a pipe in your browser, no install.